# Otter Growth Advisory — full text > Every case study on ottergrowth.com/work and every article on ottergrowth.com/notes, full text, for LLMs. Author: Michael Berliner, Product Growth Advisor & Operator (led product growth at MasterClass, built product-led growth at Calm). Otter Growth Advisory works with consumer subscription apps on conversion, onboarding, paywalls, pricing, retention, and paid acquisition. # Case studies > Anonymized engagements. Each names the product category, the business model, the timeframe, the situation, the results, and what was done. ## A CRO Program That Consistently Beats Benchmarks Source: https://ottergrowth.com/work/cro-program-beats-benchmarks Type: Case study · Product: Consumer streaming subscription · Model: Annual subscription at $180/yr · Timeframe: Ongoing, quarterly cadence Deep user research plus a high-velocity testing program lifted conversion ~26% a quarter and added roughly $30MM in incremental revenue per quarter. Results: - +26% Avg. CVR lift per quarter, net of winner's curse - ~$30MM Incremental revenue per quarter The situation: The business needed a materially higher website conversion rate to keep scaling paid marketing at a sustainable LTV:CAC. What we did: - Deep-dive purchase-journey user research and hypothesis generation - High-velocity website A/B testing - Pricing-structure testing and price-point optimization Thorough user research (on-site polling, interviews, session recordings and heatmaps, co-creation sessions) surfaced the real purchase motivators and the blockers holding buyers back. Paired with a high-velocity testing philosophy and the tooling to support it, the program produced best-in-class results quarter after quarter, lifting the revenue trajectory enough to keep scaling spend. --- ## Building B2B Product-Led Growth From Zero Source: https://ottergrowth.com/work/b2b-product-led-growth-from-zero Type: Case study · Product: Consumer meditation and sleep subscription app · Model: Consumer subscription, with a sales-led B2B team offering alongside it · Timeframe: Two years Built the B2B product-led growth product from the ground up. Self-serve conversion rose 50%, volume 100%, and B2B grew to a fifth of company revenue. Results: - +50% Self-serve conversion - +100% Self-serve volume - +40% Day-one retention The situation: There was no self-serve B2B motion at all. Every team deal, however small, went through sales. What we did: - Rebuilt the self-serve purchase flow for conversion rather than for parity with the sales motion - Raised the sales threshold from 20 to 100 employees - Smaller deals converted self-serve instead of waiting in a sales queue that was never going to prioritize them - Designed a day-one activation framework from first principles - Correlated early in-app actions against long-term retention to find which ones actually predicted it - Built the first session around reaching that action, rather than around a product tour The interesting part was not the funnel work, it was the threshold. Sales owned every account over twenty employees, which meant the deals most likely to convert on their own were the ones waiting longest for a human. Moving the line to a hundred did more for self-serve volume than any change to the purchase flow, and it cost nothing to ship. The activation framework is the piece that lasted: once you know which day-one action predicts retention, the first session stops being a tour and starts being a route to that one thing. --- ## Nearly Tripling Paid LTV:CAC in Four Months Source: https://ottergrowth.com/work/tripling-paid-ltv-cac-subscription-app Type: Case study · Product: Consumer entertainment subscription app · Model: Consumer subscription, monthly and annual · Timeframe: 4 months Moved the paywall, restructured the Meta account, and took paid LTV:CAC from 0.15 to 0.42 by doubling how many installs reach a trial at 30% lower cost per trial. Results: - 2.7x Paid LTV:CAC, 0.15 to 0.42 - 2x Install to trial start, 10% to a 20% peak - -30% Cost per trial The situation: Paid acquisition was running at roughly a sixth of breakeven, with a paywall that asked for money before the product had shown what it did. What we did: - Moved the paywall behind the product's value moment instead of ahead of it - Restructured the paid account and reallocated on evidence - Social proof and personalization copy ran 19% better on cost per result and 27% cheaper on CPM than product-description copy - Reactivated the top performer at $2.90 cost per result against a $4.46 blended average - Rebuilt the paywall offering after finding the previous change had roughly halved trial conversion on the same SKU - Redefined activation as a two-beat event and instrumented it, so the first session had a target instead of a tour Paid LTV:CAC nearly tripled, and it is worth being precise about where that came from. Not from converting trials better, which got worse over the same window, and not from cheaper installs, which got more expensive. It came from getting twice as many installs into a trial at 30% less cost per trial, which is what moving the paywall past the value moment does. That is also the honest limit of the result: at 0.42 the account is not yet at breakeven, and the next gain has to come from the trial-to-paid side rather than the top of the funnel. --- ## A Full-Funnel Audit to Hit Target LTV:CAC Source: https://ottergrowth.com/work/full-funnel-audit-target-ltv-cac Type: Case study · Product: Personal finance app for paying down loans · Model: Annual plan at $120/yr, monthly at $12.99/mo · Timeframe: 4 months Benchmarked every funnel stage against best-in-class products, then fixed the four that mattered. Average LTV rose 144% and LTV:CAC 120% in four months. Results: - +144% Average LTV - +120% LTV:CAC The situation: The growth engine wasn't acquiring customers profitably enough to scale. What we did: - Significantly increased paid-social creative velocity - Implemented rich data signals for paid-algorithm optimization - Restructured and optimized the checkout flow - Ran pricing strategy and testing - Introduced an annual plan and validated the structure across annual-only, annual + monthly, and monthly-only - Tested higher price points to lift LTV while holding conversion Benchmarking each stage of the acquisition funnel against the UX of established products made a handful of quick-win opportunities obvious: more creative velocity for paid social, richer data signals for the paid algorithms, a cleaner checkout flow, and pricing aligned to the value users actually received. Prioritizing those four drove a sharp jump in revenue trajectory and profitability. --- ## An Activation Overhaul That Doubled Activation Source: https://ottergrowth.com/work/activation-overhaul-doubles-activation Type: Case study · Product: Youth sports video and AI highlight app · Model: Consumer mobile subscription, single premium tier at $9.99/mo or $72/yr · Timeframe: 3 months Audited the signup and activation flows, cut the branching down to one core value prop, and doubled the activation rate. Results: - +109% Activation: registration to first AI movie - +43.6% Signup rate The situation: The create flow offered three ways to make something at the exact moment the product needed to prove what it did. What we did: - Audited the signup flow and the first-session activation flow end to end - Ran user interviews, which surfaced that people wanted it done for them rather than more control - Cut the branching and fronted one core value prop instead of three entry points - Moved the paywall later and collapsed the tiers to a single premium plan with an annual anchor Three ways to create something is a menu, and a menu at the aha moment is a decision the user did not ask to make. Interviews were unambiguous: they wanted the product to do it for them. Cutting the branching to one path doubled activation. The honest footnote is that trial starts fell at the same time, because the paywall moved later and a freemium gate went in, which is the argument for instrumenting activation and monetization as one system rather than two wins measured separately. --- ## An Acquisition Engine Overhaul That Lifted Leads 65% Source: https://ottergrowth.com/work/acquisition-engine-overhaul-b2b-saas Type: Case study · Product: B2B team-management SaaS for small and mid-size companies · Model: $10 per user per month, self-serve, no sales team · Timeframe: 4 months Stood up paid acquisition from zero, rebuilt the conversion funnel from landing page to account creation, and instrumented all of it. Weekly leads rose 65% and signup completion doubled. Results: - +65% Weekly new-account leads - 2x Signup-to-account-creation The situation: No paid acquisition, a marketing site that could not be measured end to end, and no way to tell whether a signup ever became a team that used the product. What we did: - Stood up paid acquisition from zero: picked the channels, built the first ad creative on five tested motivators, set budget allocation across Google and Meta - Rebuilt the conversion funnel: migrated the site to an AI-native platform, rebuilt the landing pages and a competitor comparison hub, reworked account creation in two releases. Completion doubled - Instrumented it end to end: one event taxonomy from site visit to account to first teammate invite, read weekly - Checked the numbers before chasing them: a reported halving of signup-to-trial was a counting error, an organic decline was branded click-through, not rankings. Both got a hold, not a project - Moved the goal to activated accounts: defined the metric with leadership, wrote the events engineering needed, designed the activation and lifecycle program to move it - Handed over a repo the team runs: every deliverable, a timeline, open questions with owners, and the working methods Four months in, the easy read was that acquisition had worked and the job was to push harder. Leads were up 65%. But new revenue had not followed, and an account that never invites a teammate is not a customer. So the work shifted to defining an activated account, getting the exact events in front of engineering, and building the program to move that number. The overhaul is the headline. The shift is what mattered. --- ## Scaling Meta Ads While Increasing Spend Efficiency Source: https://ottergrowth.com/work/scaling-meta-ads-spend-efficiency Type: Case study · Product: At-home health testing subscription · Model: Annual subscription at $180/yr · Timeframe: 8 months Restructured targeting and consolidated onto winning creative to cut CPC 26% while scaling spend 260%. Results: - +260% Ad spend scaled - +26% CPC efficiency The situation: Clicks cost too much for the paid engine to pay for itself, so the account had been held small. Spend could not go up without LTV:CAC going down. What we did: - Consolidated fragmented targeting into a few broad ad sets - Budget was split across enough audiences that no single ad set gathered the conversion volume Meta needs to optimize against - Fewer and broader lets the algorithm find the buyer rather than being told where to look - Moved budget onto the winning creative and kept iterating on it - Creative carries the targeting now: a different message per ICP is the signal the algorithm acts on - The territories were built on value props users had voted for, not on what the team wanted to say - Scaled in steps, reading cost per link click at every one - Spend tripled by month two, came back down through the spring, then ran at three to five times the starting level from month six on - Cost per link click fell from $1.60 to $0.85 at its cheapest, and held at $1.20 with spend near its peak Scaling spend usually costs you efficiency. Here it went the other way: a cleaner targeting structure and relentless consolidation onto the creative that worked let spend grow 260% while CPC actually improved. --- ## Overhauling the Acquisition Engine for a Profitable LTV:CAC Source: https://ottergrowth.com/work/overhauling-acquisition-engine-ltv-cac Type: Case study · Product: Seed-stage education streaming subscription · Model: Annual subscription at $180/yr, monthly tier removed · Timeframe: 8 months Pricing, on-site conversion, and cleaner paid-algorithm signals took LTV:CAC up 43% and to breakeven in eight months. Results: - +37% Average LTV - +43% LTV:CAC - 1:1 Reached breakeven LTV:CAC The situation: The acquisition engine wasn't running at a profitable LTV:CAC. What we did: - Ran pricing strategy and testing - Removed the monthly price tier - Optimized the core price point - Website conversion rate optimization - Optimized the data signals feeding the paid algorithm --- ## Building a CRO Experimentation Program From Zero Source: https://ottergrowth.com/work/cro-experimentation-program-from-zero Type: Case study · Product: Managed marketplace for refurbished e-bikes · Model: Buys, refurbishes, and resells e-bikes across three European markets · Timeframe: 3 months Stood up an A/B experimentation program from nothing and ramped it to four experiments every two weeks, with the operating model owned by the client's team. Results: - 0 to 1 CRO experimentation program built - 4 Experiments shipped every two weeks The situation: A growth-stage marketplace with no experimentation capability, where the binding constraint was supply rather than demand. What we did: - Framed a conversion brief as a supply problem and descoped the demand storefront - Implemented A/B experimentation tooling (GrowthBook) alongside a component library, so tests could be built and shipped without bespoke engineering each time - Wrote the operating model the team runs - Experiment build guide and results guide - Pre-sprint ticket audit cadence - A definition of ready, so nothing enters a sprint half-specified - Set supply-side guardrails distinct from the demand funnel's, and held that line when seller-flow tests were judged on buyer metrics - Ran a competitive teardown across car, electronics, and bike resale marketplaces to source the hypothesis backlog - Reconciled two conflicting offer-acceptance rates into one number for the CEO Most teams that want a testing program buy a tool and stall, because the tool was never the constraint. The constraint is a backlog worth testing, a build path that does not need bespoke engineering per test, and an agreed bar for what counts as a win. Those three got built alongside the tooling, which is why the cadence held at four experiments every two weeks rather than spiking and dying. Early results are landing now and the compounding read comes at the end of the program. --- ## A CRO Audit of an Employer Onboarding Funnel Source: https://ottergrowth.com/work/employer-onboarding-cro-audit Type: Approach · Product: Global online job marketplace · Model: Two-sided marketplace, employers pay to post and promote roles · Timeframe: One month An annotated audit board: where the revenue gap sits, a nine-platform teardown, ICE-scored quick wins each with a control and a variant, and a prioritized roadmap. The situation: They wanted an outside read on the employer funnel, from landing page to first job posted, and a roadmap someone could actually execute. The approach: - Seven observations on the current flow, each anchored to where the revenue gap actually sits rather than to what looked worst - A nine-platform competitive teardown - Direct competitors alongside adjacent paid-listing products, because the patterns that matter are often outside the category - Eight ICE-scored quick wins, each with a control and a variant - Written so a designer or engineer could pick one up without a meeting first - Six prioritized recommendations, each carrying the reasoning rather than just the verdict - A backlog of A/B test hypotheses to run after the quick wins land An audit that names problems is easy and mostly useless, because the team already knows where it hurts. What they cannot do from the inside is rank the fixes against each other and say what to do on Monday. So every observation here carries a sized opportunity, every quick win ships as a control and a variant so the result is readable, and the teardown ranges wider than the obvious competitors, since the strongest patterns in an employer funnel turned out to be sitting in local-services advertising rather than in job boards. --- ## A Growth Dashboard That Explains Its Own Numbers Source: https://ottergrowth.com/work/growth-pulse-dashboard-that-explains-itself Type: Approach · Product: Consumer entertainment subscription app · Model: Consumer subscription, monthly and annual · Timeframe: 4 months Built a growth dashboard where every metric carries the benchmark, the current value, the cause, and the move. Two of the headline numbers were actively misleading without it. The situation: The team had metrics and no read on them, so the numbers that mattered most were the ones being misinterpreted. The approach: - Gave every metric four fields instead of one: the benchmark, the current value, what is driving it, and the opportunity - Benchmarked against the right access model - The paywall sat inside signup with a dismissible exit, so the number had to be read as a subscription product with a leaking gate rather than as a free product - Picking the wrong comparison set moved expected conversion by several multiples in either direction - Named the two metrics that mislead without context - Blended cost per install, because broken attribution hides paid installs inside the organic number - Retention, which reads as a product failure but is a funnel failure, because it counts users who never reached the core against the product - Found a paywall offering change that had quietly halved trial conversion on the same SKU - Reduced it to five plain-language insights an exec could act on without a translation layer Most growth dashboards answer what happened and stop, which leaves the reading to whoever opens it first. That is where "paid is dead, organic is booming" comes from, when the truth was that paid installs were hiding inside the organic number because attribution in this category is broken. Every metric here carries our read of the cause, so the dashboard argues rather than just reports. The most valuable thing it surfaced was not a new opportunity, it was a change that had already happened and gone unnoticed. --- ## Designing a Free-to-Paid Transition Without the Churn Spike Source: https://ottergrowth.com/work/free-to-paid-transition-without-the-churn-spike Type: Approach · Product: Operating-system SaaS for small and mid-size companies · Model: $10 per user per month, self-serve · Timeframe: 4 months Designed a reverse trial for 6,700 existing free users around the failure case rather than the success case, and redefined activation as a multi-user event. The situation: 6,700 people already had the product for free, and moving them to paid risked the churn spike that usually follows a takeaway. The approach: - Designed the transition as a reverse trial rather than a takeaway - Built against the cautionary case: a 44% churn spike after a transition with no grandfathering and a week's notice - Modeled on the version that worked: a genuinely useful solo free tier, with only team features restricted - Modeled conversion at 5-18% depending on how much of the base activates as teams - Wrote the comms sequence from 60-90 days out through the post-trial offer - Redefined activation as multi-user - Solo users will not convert at any trial structure - The real activation event is inviting a teammate before the trial expires - The category pattern: roughly 90% of workspace creators in collaboration tools never invite anyone - Killed a paid channel on unit economics - A $400 blended CAC is structurally loss-making below 5 seats per account - At one seat it is 0.5:1 LTV:CAC - Ruled a marketing angle off-limits because it was a product problem, not a message problem Most free-to-paid transitions fail on the same mechanic. Existing users read the change as something being taken away, because it is. So the design question is not what to charge, it is what to add so the trial reads as additive. Activation is the same problem one step earlier: a single user evaluating team software alone will not convert no matter how the trial is structured, which makes the teammate invite the event worth designing everything around. --- ## Building a User-Centric Ads Strategy Source: https://ottergrowth.com/work/user-centric-ads-strategy Type: Approach · Product: At-home health testing subscription · Model: Annual subscription at $180/yr · Timeframe: 1 month Polled real users to find what they valued most, then built each ad-creative territory directly on a validated value prop instead of guessing. The situation: Finding an ad-creative strategy that genuinely resonated with the product's core user segments. The approach: - Polled users directly to identify core needs ("Which of these is most valuable to you?") - Identified the key user value props from the responses - Cost savings over the alternatives - Achieving specific health goals - Doctor or trusted-expert approval - Solving a specific, urgent need - Translated each validated value prop into its own ad-creative territory, hook, and messaging framework Most ad-creative strategy starts from what the team wants to say. This started from what users said they valued. Direct polling turned a guessing game into four distinct creative territories, each anchored to a value prop we already knew resonated, ready to test with real spend. --- ## Auditing a Paid Funnel With Nothing Measuring It Source: https://ottergrowth.com/work/paid-funnel-audit-with-no-tracking Type: Approach · Product: iOS journaling app · Model: Digital subscription at $4.99/mo or $49.99/yr, plus a mailed-letter tier at $19.99/mo · Timeframe: 2 weeks The ad account had run 325 days with zero events recorded. The audit reframed the problem from overspending to never having spent or measured enough to learn. The situation: A solo founder was spending on paid acquisition with no way to know what was suppressing installs and paid conversion. The approach: - Audited all four stages: paid ads, store listing, signup, paywall - Opened the ad account first, and found the pixel and SDK had never been installed - Zero events in 325 days - Campaigns were optimizing to landing-page views with no post-click signal - $131.86 of total spend across 11 months, so nothing ever reached a learning phase - Benchmarked against the right profile - Paywall-first rather than freemium, so install-to-paid benchmarks at ~11%, not the ~2% freemium median - Mixing freemium stage rates with hard-paywall endpoints would have set the wrong target - Ranked the fixes: tracking first, then paywall structure, then the store listing, then onboarding sequence The founder came in expecting to hear his ads were too expensive. Cost per install was being quoted in the hundreds of dollars. That number was never real, because with no pixel and no SDK there was never an install event to measure against. The raw material was fine: the best creative was pulling 11-13% CTR at about $0.11 a click. Everything after the click was invisible. The first fix was not a bid or a creative, it was installing the thing that lets you learn. --- # Articles ## Education app retention playbook: onboarding, the first 48 hours, and the goal loop Source: https://ottergrowth.com/notes/education-app-retention-playbook-onboarding-activation Published: 2025-11-24 > You are not competing with other learning apps. You are competing with TikTok, and day-30 retention for education sits around 3 to 4%. You cannot change that equation, but you can tilt it: commitment at onboarding, aggressive activation inside the first 48 hours, and a goal-loop screen at the end of every session. Make perceived value dwarf perceived effort. Education app retention is one of the hardest problems in consumer tech. And most teams don't even understand why they're losing. You're not just competing with other learning apps. You're fighting Netflix, TikTok, and every other low-effort dopamine hit for your users' time. Learning demands high cognitive load and sustained motivation. Entertainment requires neither. This asymmetry defines everything about your retention strategy, and it shows up in the benchmark. Day-30 retention for education apps sits around 3 to 4%. That is not a broken category, that is the category. If you are measuring yourself against a social app's numbers you are going to fire the wrong people. After advising multiple education apps, from language learning platforms to theological training programs to wellness apps, I've seen this unfair fight play out repeatedly. The apps that survive understand one critical truth: you can't change the motivation equation, but you can tilt it dramatically in your favor. It boils down to making the perceived value >> the perceived effort. [Figure: Retention is won by making perceived value far exceed perceived effort. The perceived-value bar is long; the perceived-effort bar is short. The gap between them is where retention is won.] You cannot make learning effortless. You win by making the value dwarf the effort: commitment at onboarding, fast activation, and a visible goal loop. Here's the playbook that actually works for improving education app retention. ## Onboarding Is Where Motivation Lives or Dies Most apps treat onboarding as product education. Show users where the buttons are. Explain how the features work. Maybe throw in a quick demo lesson. That's a mistake. Onboarding is your singular opportunity to get users to commit, not just understand but genuinely commit, to their learning journey. There's a massive difference between "I get how this works" and "I'm invested in making this work." So your onboarding needs to focus on building buy-in, not just comprehension. ### Three Tactics That Drive Real Commitment **1\. Explicit Goal Setting** Don't just ask what they want to learn. Make them state specific, measurable intentions. "How many hours per day do you want to learn?" "What level of fluency are you targeting?" "When do you want to achieve this goal?" The act of articulating specific goals creates psychological investment. Users who state clear intentions follow through far more often than users who vaguely "want to learn Spanish." **2\. Opt-In Accountability** Once they've stated their goal, immediately offer them the tools to stay accountable. "Want a daily reminder to keep you on track?" "Should we check in with you if you miss a day?" The key word here is opt-in. When users choose the nudge themselves, it's not annoying, it's supportive. They've given you permission to hold them accountable. This small shift dramatically improves retention because users feel ownership over their learning path. **3\. Challenge-Based Commitment** Appeal to their competitive side by framing learning as a challenge they're accepting. "What streak do you want to hit this week?" "Can you complete 5 lessons in the next 3 days?" Duolingo absolutely nails this with their streak mechanics. But the principle applies beyond gamification. Humans respond to challenges they've accepted. Make them accept one during onboarding. ### Why Most Apps Get This Wrong Most apps only implement one of these tactics. Some do goal setting and accountability. Almost none combine all three to create multiple, reinforcing commitment mechanisms. You're not just setting expectations, you're building psychological investment before they've completed a single lesson. That investment becomes the foundation for everything that follows. ## The First 48 Hours: Make or Break for Education App Retention Here's the harsh reality: if users don't experience deep value in the first 24-48 hours, they're gone. They'll ghost you for Instagram, their friends, or literally anything easier than learning. The window for proving your value is brutally short. This is why early activation is the most critical lever for retention. ### Aggressive Value Delivery Your job during this window is simple but requires aggressive execution: push them to complete as many lessons as possible before competing distractions win. Not to overwhelm them. Not to be annoying. But to frontload the "aha moment" before they forget why they downloaded your app in the first place. This requires communication intensity that would be excessive for most apps. I recommend three touchpoints per day across email and push during those first two days. Yes, three per day. Morning motivation, midday reminder, evening encouragement. Mix the channels. Vary the messaging. But maintain presence. ### Why Over-Communication Works Most product teams balk at this approach. "That's too much. Users will be annoyed. They'll uninstall." Maybe. Some will. But here's what I've seen across multiple education app engagements: the cost of under-communicating, losing users who never experienced your value, is far higher than the cost of over-communicating to a small subset who find it excessive. According to Reforge's research on activation metrics, users who complete 3+ core actions in their first session have 4x higher retention than those who complete just one. Your aggressive communication during those first 48 hours exists to drive those critical early completions. The users who unsubscribe from your emails on day 2? They were never going to become retained users anyway. Don't optimize for them. Instead, optimize for the users who need that extra nudge to complete one more lesson, experience one more win, and cross the threshold from "trying this app" to "this app is working for me." ## Close the Goal Loop: The Education App Retention Strategy No One Executes Well Once you've got users activated, the standard playbook is streaks, nudges, and habit-building mechanics. That's table stakes. Duolingo does it. Memrise does it. Every education app does some version of streaks and daily reminders. What separates great education apps from mediocre ones is relentlessly closing the goal loop. This is the most underutilized tactic in retention. ### Make Progress Visible Users don't feel progress unless you explicitly show them. This isn't obvious to product teams, but it's brutally true. Think about it: when you're learning something difficult, the progress feels invisible day-to-day. You completed a lesson. Great. But did that actually move you closer to fluency? To your goal? Who knows? This ambiguity kills motivation. Your job is to make progress visible, tangible, and directly connected to the goal they set during onboarding. ### How to Close the Loop After every session, in every email, throughout your app: remind them of (1) their original goal, and (2) how much progress they've made toward it. "You wanted to reach conversational Spanish in 6 months. You're now 15% of the way there." "Your goal was 30 minutes of learning per day. You've hit that target 12 days in a row." "You set out to complete the beginner track. You've finished 8 of 20 modules." Personalize it to their specific commitment whenever possible. Generic progress bars don't cut it. Connect their current state back to their stated intention. ### Build it as a screen, not a message The thing to actually build is the last screen of a session. Three elements, in this order: the number they gave you at onboarding, where they are against it right now, and the one next step that closes a little more of the gap. "3 of 4 done this week" beats any streak flame, because the denominator is theirs. A streak is a number your product invented. A goal is a number they said out loud, and only one of those survives a bad week. That screen is the last step of the playbook and the one that gets dropped, because it is the only one that is not a feature. It ships no new content, no new lessons, no new mechanic. It just points at what the user already told you, which is why it never wins a roadmap argument and why it moves retention more than the things that do. [Figure: The goal loop: set a specific goal at onboarding, make progress, show that progress against the goal, and motivation is renewed, which drives more progress.] Every session is a chance to re-sell users on their own decision: connect their progress back to the goal they set at onboarding. ### The Impact on Education App Retention This single tactic can deliver 2-3x improvement in retention. I've seen this play out across multiple clients. Apps that implement consistent goal-to-progress messaging see dramatic improvements in D7, D30, and D90 retention. Users stick around because they can actually see they're getting somewhere. Nir Eyal's Hook Model makes the same point: variable rewards combined with investment drive habit formation. Closing the goal loop provides that reward signal users need to keep coming back. ## The Innovation Layer: Making Hard Things Feel Easier Let's be honest: learning will always require cognitive effort. You can't change that fundamental reality. But you can change how that effort feels. The most innovative apps are pushing the boundaries here by making learning feel less like work. ### Three Approaches That Work **Game Mechanics That Create Flow States** Duolingo's genius isn't just streaks, it's how they've made lessons feel like playing a game rather than studying. Quick wins. Immediate feedback. Variable rewards. The core learning is still happening, but it's wrapped in mechanics that trigger dopamine. **AI Agents That Reduce Friction** Language learning apps are starting to use AI conversation partners that adapt to your level in real-time. Instead of rigid pre-scripted lessons, you're having an actual conversation that naturally pushes you just beyond your current ability, the optimal zone for learning. **Stupid-Easy Daily Integration** The best education apps make it stupid-easy to maintain momentum. Five-minute lessons. Offline access. Integration with your routine rather than requiring you to carve out dedicated study time. ### You Don't Need Sophistication to Win These innovations matter. They tilt the unfair motivation equation slightly back in your favor. But here's what I've learned: you don't need these sophisticated features to succeed at retention. They're not the foundation, they're the acceleration layer. The foundation is still commitment during onboarding, aggressive activation in the first 48 hours, and relentless goal-to-progress alignment. Get those three right, and you can compete. Add innovation on top, and you can dominate. ## Why This Education App Retention Playbook Works The fundamental problem with education apps is that learning is hard and entertainment is easy. You can't change that. No amount of gamification or AI tutoring changes the fact that acquiring a new skill requires sustained cognitive effort over weeks and months. What you can change is whether users remember why they signed up in the first place. Most education apps lose users not because the learning experience is bad, but because users forget their original motivation. Life gets busy. Other apps are easier. The initial excitement fades. That's why commitment during onboarding matters. That's why aggressive activation in the first 48 hours matters. That's why relentlessly closing the goal loop matters. You're not just teaching users Spanish or coding or how to run faster. You're constantly re-selling them on their own decision to learn. Every lesson completion, every email, every push notification is an opportunity to remind them: "You set out to do this. Look how far you've come. Keep going." When users can see they're actually progressing toward something they care about, the hard work starts to feel worth it. That's how you win the unfair fight for education app retention. ## Your Next Steps If you're building or growing an education app, here's what to implement first: **Week 1: Audit your onboarding** Do you get explicit goal setting? (Not just "What do you want to learn?" but measurable, specific goals.) Do you offer opt-in accountability mechanisms? Do you frame learning as a challenge users are accepting? **Week 2: Intensify your activation window** Track how many lessons users complete in their first 48 hours. Increase communication frequency during this window (test 3 touchpoints per day). Measure the impact on D7 retention. **Week 3: Close the goal loop everywhere** Add goal-to-progress messaging after every session. Include it in every lifecycle email. Surface it prominently in your app's home screen. The motivation equation for education apps will always be unfair. But with the right playbook, you can tilt education app retention far enough in your favor to build a thriving learning product. ## Common questions ### Why do education apps struggle to retain users? Because they compete with entertainment for the same time, and learning is high-effort while entertainment is not. Day-30 retention for the category sits around 3 to 4%, which is the benchmark rather than a sign something is broken. Most users churn not because the content is bad, but because they forget the motivation they had when they signed up. ### What is the most underused education app retention tactic? Closing the goal loop: after every session, connecting the user's progress back to the specific goal they set at onboarding. Done consistently, it can deliver a 2-3x improvement in retention. ### How aggressive should onboarding communication be? In the first 48 hours, more aggressive than feels comfortable. Around three touchpoints per day across email and push, to drive the early lesson completions that predict long-term retention. The users who opt out that early were rarely going to stick anyway. --- ## Referral programs as a growth channel: a framework for realistic expectations Source: https://ottergrowth.com/notes/referral-growth-channel-framework Published: 2025-11-21 > A well-run referral program adds about 10-20% incremental acquisition (a k-factor of 0.1-0.2), not viral growth. Build the three levels in order: incentivized referrals (table stakes), content sharing (the multiplier), and viral in-app features (rare, and only if your product's core is social). Referrals improve your unit economics; they do not replace paid. The piece of the mix that actually changes how much paid you have to buy is offline word of mouth, and you earn that with a delightful product rather than a program. ## The Referral Expectation Problem After working with 12 B2C subscription apps over the last 18 months, I keep seeing the same pattern: founders expect referrals to solve their CAC problems. They launch a "give $20, get $20" program and wait for the hockey stick growth curve. It doesn't come. Not because their program is broken. But because their expectations are completely detached from reality. Here's what I tell every client when they ask about referrals as a growth channel: a well-executed referral program typically contributes 10-20% incremental acquisition. In technical terms, that's a K factor of 0.1 to 0.2. That's good. That's worth building. But that's not the explosive viral growth most founders are chasing. Here is what actually works, what doesn't, and how to think about referrals strategically. ## Where Referrals Sit in Your Acquisition Mix Before the framework, the mix. Referrals are one slice of it, and knowing how big that slice can get is what makes the rest of this article usable. **Paid does most of the work.** For almost every consumer product that's normal, and it's what the channel is for. **Referrals, 10 to 20%.** That is incentivized referrals plus content sharing together, and unless your product is genuinely easy to share, almost all of it comes from the incentivized side. **Organic, meaning social and SEO, another 10 to 20%,** and it takes 8 to 12 months before it shows up at all. **Offline word of mouth.** This is the X factor, and it's a completely different thing from a referral program. Nobody clicked a link. Somebody told a friend. [Image: Two acquisition mixes side by side. For a product people do not talk about, paid carries almost all of it forever, with referrals at 10 to 20 percent and organic social and SEO at 10 to 20 percent. For a product people talk about, the same referrals and the same SEO sit alongside a large offline word of mouth slice, and paid does about a third of the work.] Same referral program, same SEO, two very different businesses. Word of mouth is the only slice that changes the size of the paid slice. Look at those two mixes and the point of the whole article falls out of it. Referrals and SEO cap out wherever they cap out. Doubling down on a referral program moves a slice that was never going to be more than a fifth of your acquisition. The thing that changes the shape of the business is the slice you can't buy. ## Offline Word of Mouth Is the X Factor Word of mouth decides how much paid you have to buy. That's the whole reason it matters more than the program you're about to build. You can't buy it, and you can't run it as a campaign. Three things have to be true before you get any of it: **The product has natural word-of-mouth potential.** It's the kind of thing that comes up in conversation. Plenty of products aren't, and no amount of execution changes it. **People are genuinely excited about it.** Not satisfied. Excited enough to bring it up unprompted, which is real product-market fit rather than a decent retention curve. A product can hold its users and still give nobody a reason to mention it. **The core product is delightful**, not just fine. Fine gets you retention. Delightful gets you told about. A lot of products don't have that potential to begin with, and plenty of the rest aren't delightful enough to earn it. That's worth being honest about before you budget a quarter against it, because the answer changes what you work on. If word of mouth isn't available to you, paid is carrying this business for the foreseeable future, and product work is the only thing that changes that. So build the referral program, take your 10 to 20%, and stop waiting for it to fix acquisition. The work that changes your mix is making the product worth talking about. ## The Viral Growth Myth According to Reforge's benchmarking data on consumer apps, products with K factors above 0.5, meaning each user brings in at least half a new user through sharing, are exceptionally rare. We're talking less than 1% of apps. And when you look at what those apps actually are, a pattern emerges immediately: social media platforms, communication tools, collaboration software. Products like ChatGPT, Instagram, Slack, or Discord. Products where the core value proposition is inherently social or multiplayer. The reason? These products have sharing built into their DNA. You literally cannot get full value from Instagram without your friends being on Instagram. ChatGPT conversations became shareable status symbols. Discord's entire model is based on inviting people into communities. Most products don't have this luxury. If you're building a fitness app, a meditation app, a language learning tool, or a productivity tracker, your product delivers value to individual users. Sharing might enhance the experience, but it's not core to the value proposition. That doesn't mean referrals don't matter for these products. It just means you need a different framework. ## The Three-Level Referral Framework Over the years, I've developed a framework for thinking about referral strategies across three distinct levels. Not every product can, or should, operate at all three levels. But understanding where your product fits helps you set realistic goals and allocate resources appropriately. ### Level 1: Incentivized Referrals (The Table Stakes) This is the classic referral program: give $20, get $20. Or give a free month, get a free month. Or refer three friends and unlock lifetime premium access. The mechanics are straightforward. When an existing user refers a new user who completes a desired action (usually making a purchase or completing onboarding), both the referrer and the referred get some reward. **Why this works:** Incentivized referrals create a clear value exchange. You're essentially paying your existing users a customer acquisition cost in the form of credits, discounts, or perks. And you're giving new users a discount to reduce friction at signup. From an economics perspective, this often makes sense. If your blended CAC through paid channels is $40, offering a $20 credit to both parties ($40 total cost) to acquire a customer can be cost-neutral while improving your customer quality. Referred customers typically have higher retention rates and LTV than paid acquisition. **How to execute this well:** First, make the offer valuable enough to motivate action. A $5 credit when your product costs $50/month won't move the needle. You need the incentive to feel meaningful. Second, make sharing frictionless. One-tap sharing to text, email, or social with pre-written copy performs significantly better than making users hunt for a referral link buried in settings. Third, remind users regularly. Most users won't think to refer friends unless you prompt them. In-app notifications after positive moments (completing a goal, finishing a milestone, etc.) convert better than random reminders. At an education app I advise, we implemented an incentivized referral program that now drives approximately 15% of new user acquisition. Not revolutionary, but that's 15% of users coming in at significantly better unit economics than paid social. **The baseline expectation:** A solid incentivized referral program should deliver 10-15% incremental acquisition. If you're getting less, something's broken in your execution. If you're getting more, congratulations, you've likely built a product people genuinely love sharing. [Figure: The three-level referral framework as ascending tiers. Level 1 incentivized referrals adds 10 to 15 percent. Level 2 content sharing adds another 5 to 10 percent. Level 3 viral in-app features can reach a k-factor of 0.5 to 0.8 but is rare and needs a social product core.] The three levels, by leverage and by how rare they are. Build them in order; Level 3 is a bonus most products never earn. ### Level 2: Content Sharing (The Multiplier) Content sharing focuses on enabling users to share their accomplishments, achievements, or helpful content directly from your app. This is Strava letting you share your run. Duolingo celebrating your streak. Headspace showing your meditation minutes. Any app that generates shareable moments or results that reflect well on the user. **Why this works:** Content sharing taps into natural human behavior. People want to share their progress, celebrate wins, and demonstrate expertise or consistency. You're not asking them to shill your product, you're giving them a way to showcase their own achievements. Content shares also reach beyond a user's immediate network. A great workout result shared on Instagram reaches hundreds of people who might not know the sharer personally but are inspired by the content. **How to execute this well:** The content must make the sharer look good. No one wants to share something that makes them look bad or mediocre. That's why Strava highlights personal records and milestones, not average Tuesday runs. Make the share visually compelling. A text-only share won't cut it. You need beautiful graphics, clear data visualization, or aesthetically pleasing layouts. Think about how the share will look in a social feed. Add automatic branding without being heavy-handed. Your logo and app name should be visible but not dominating the image. The focus is the user's achievement, not your marketing. For a consumer fitness client, we redesigned their post-workout share cards to be more visually striking and added automatic comparison to previous bests. Share rate increased by 40%, which translated to roughly 8% incremental acquisition from these organic social shares. **The baseline expectation:** Content sharing typically adds another 5-10% incremental acquisition on top of your incentivized program. This depends heavily on whether your product naturally creates shareable moments. Meditation apps and fitness apps have an easier time here than, say, password managers. ### Level 3: Viral In-App Features This is where real leverage lives, and where most products can't play. Viral in-app features are product mechanics that naturally inspire sharing or create talk-worthy moments. These aren't tacked-on share buttons. These are features where sharing is embedded in the core user experience. Examples: - Spotify Wrapped (everyone shares their year-end data) - Wordle (sharing your score became part of the game) - Loom (recipients become users when they reply with their own video) - Notion (collaboration requires inviting others into your workspace) **Why this works (when it does):** Viral features create K factors of 0.5 to 0.8 or higher because they make sharing either necessary for the product to function or incredibly compelling for the user experience. The key difference: these features don't feel like marketing. They feel like product. Users share because it enhances their experience or completes a workflow, not because you offered them $20. **The brutal truth about viral features:** Most products can't build them. Your product needs to have an inherent social component, competitive element, or collaborative workflow to support viral features. You can't force virality onto a fundamentally single-player experience. I've seen too many teams waste engineering resources building "viral features" that no one uses because they were bolted onto a product that didn't naturally support them. A budgeting app with a "share your savings goals" feature isn't viral, it's just another ignored button in the UI. **When to invest in viral features:** Ask yourself: Is there a core workflow in my product that would be genuinely enhanced by social interaction, collaboration, or friendly competition? If yes, develop strong hypotheses and test them rigorously. Start small. Spotify didn't launch with Wrapped, they built it years into their product's life when they had the data and audience to make it impactful. If no, don't force it. You're better off perfecting Levels 1 and 2 and investing your engineering resources into improving core product value. [Figure: Viral in-app features work only when the product core is social, multiplayer, competitive, or collaborative, like Spotify Wrapped, Wordle, Loom, and Notion. Single-player products such as meditation, fitness, and budgeting should stay at Levels 1 and 2.] You cannot force virality onto a single-player product. Be honest about which side you are on before you spend engineering time. ## How to Implement This Framework Here's my recommended approach for most companies: **Start with the table stakes.** Build a solid incentivized referral program. Make the reward compelling, the sharing frictionless, and the prompts well-timed. Get this to the baseline 10-15% acquisition contribution. **Add content sharing if applicable.** If your product creates achievements, milestones, or impressive results, enable users to share them beautifully. Invest in the design of these share cards, they're marketing materials that your users create for you. **Evaluate viral features honestly.** Once Levels 1 and 2 are working, assess whether your product has the DNA to support viral mechanics. Be brutally honest. Most won't. If you do see potential, develop hypotheses, prototype cheaply, and test with a subset of users before committing significant engineering time. **Optimize relentlessly.** Like any growth channel, referrals require ongoing optimization. Test reward amounts, messaging, timing, share card designs, and distribution channels. Small improvements compound. ## The Real Value of Referral Programs Back to reality. Referrals probably won't transform your business overnight. They won't replace paid acquisition. They won't solve your product-market fit issues. What they will do is improve your blended CAC, increase your LTV:CAC ratio, and fuel your organic growth engine. That 10-20% incremental acquisition at better unit economics is valuable. Over time, it compounds. Referred users also often have higher retention rates and better engagement than paid acquisition. They're coming in with social proof, a friend or acquaintance recommended the product. That matters. Every company should have at minimum Levels 1 and 2 implemented and executed well. These are table stakes for modern consumer products. If you're not doing them, you're leaving growth on the table. But keep your expectations grounded. Unless you successfully build viral in-app features (and again, most products can't or shouldn't), referrals are a supplement to your paid acquisition strategy, not a replacement. ## Common Mistakes to Avoid Over the years, I've seen teams make the same referral program mistakes repeatedly. Here are the ones to watch out for: **Mistake 1: Launching and forgetting.** Building a referral program is the beginning, not the end. You need to actively promote it, optimize it, and drive adoption. A referral link buried in settings that no one knows about won't drive results. **Mistake 2: Making the incentive too small.** If your product costs $50/month and you're offering a $5 credit, no one will bother referring friends. The juice isn't worth the squeeze. Make the reward meaningful enough to inspire action. **Mistake 3: Ignoring fraud.** Incentivized programs attract fraudulent referrals, people creating fake accounts to claim rewards. Build in fraud detection from day one, or you'll waste money on non-real users. **Mistake 4: Forgetting the user experience.** Don't interrupt critical workflows with referral prompts. Timing matters. Ask for referrals after positive moments (completing a goal, finishing a workout, achieving a milestone), not during onboarding or when users are trying to complete tasks. **Mistake 5: Forcing viral features.** As mentioned earlier, not every product can support viral mechanics. Don't waste engineering time building features no one will use. ## Measuring Success How do you know if your referral program is working? Track these metrics: **Referral rate:** What percentage of active users are referring at least one person? Benchmark: 20-30% is solid for most consumer apps. **Conversion rate:** What percentage of referred users actually sign up and convert to paying customers? This should be higher than your paid acquisition conversion rate, if it's not, something's wrong with either your targeting or your reward structure. **K factor:** (# of invites per user) × (conversion rate of invites) = K factor. As discussed, 0.1 to 0.2 is realistic for most products. **LTV of referred users:** Compare the lifetime value of referred customers versus paid acquisition customers. Referred users should have higher LTV due to better retention. **Referral CAC:** Calculate the total cost of your referral program (rewards paid out) divided by customers acquired through referrals. Compare this to your paid acquisition CAC. ## The Bottom Line Referrals are a valuable but realistic growth channel. Set your expectations appropriately: 10-20% incremental acquisition is success for most products. That's meaningful growth at better economics than paid acquisition, but it's not going to 10x your business. Build the table stakes, incentivized referrals and content sharing, and execute them well. Then honestly evaluate whether your product can support viral features before investing significant resources. Done right, referrals improve your unit economics, increase customer quality, and create a sustainable organic growth engine. That's worth building. But it is not the thing that decides how much paid you have to buy. Referrals and SEO cap out wherever they cap out, and doubling down on either moves a slice that was never going to be more than a fifth of your acquisition. The only slice that changes the size of the paid slice is offline word of mouth, and you get that by building something people are delighted enough to mention, or you do not get it at all. **Next step:** Audit your current referral program (or lack thereof) against this framework. Where are the gaps? Start with Level 1, nail it, then move up the stack. Then go and break your own acquisition down honestly and find out where it actually comes from. ## Common questions ### How much growth should a referral program realistically drive? For most consumer subscription apps, about 10-20% incremental acquisition, which is a k-factor of 0.1 to 0.2. That is worth building for the better unit economics, but it is not the viral growth founders often expect. ### What kinds of products can actually go viral? Ones whose core is social, multiplayer, competitive, or collaborative, like Spotify Wrapped, Wordle, Loom, or Notion. If your product delivers its value to one user on their own, do not force virality; master incentivized referrals and content sharing instead. ### Is word of mouth the same as a referral program? No, and the difference decides your acquisition mix. A referral program is a mechanic you build, and it caps out around 10 to 20% of acquisition. Offline word of mouth is people telling each other about the product. It's the only piece that changes how much paid you have to buy, and you earn it by having a product with natural word-of-mouth potential that people are genuinely excited about and that's delightful rather than fine. ### Are referred users worth more than paid ones? Usually yes. They tend to have higher retention and LTV because they arrive with a friend's recommendation as social proof, which is why referrals improve your blended CAC even at modest volume. --- ## AI-Enabled Couples Therapy: The Massive Market Opportunity Everyone's Missing Source: https://ottergrowth.com/notes/ai-couples-therapy-market-opportunity Published: 2025-11-12 > People spend thousands on individual therapy and avoid couples therapy entirely, held back by stigma and cost. AI changes the economics, making real relationship expertise accessible to the vast middle that traditional therapy prices out. That gap is one of the biggest underserved opportunities in mental health. Your relationship with your spouse is the single biggest factor affecting your mental health. Not your job. Not your finances. Your partner. But almost nobody gets help with it. Not because they don't care, but because the market has completely failed them. I just took on a new client in the AI-enabled couples therapy space, and it's opening my eyes to one of the biggest underserved opportunities in mental health. ## The couples therapy market is broken Here's what shocked me when I started digging into this space: couples therapy is one of the most stigmatized, underutilized, and antiquated categories in mental health. The stigma is real, and it is very specific: "If we need couples therapy, our relationship must be failing." But that's backwards. Couples therapy isn't just for relationships in crisis. It's valuable whether you're thriving or struggling, 6 months in or 16 years in. You know, you don't wait until you're obese to start exercising. Why wait until your relationship is broken to invest in it? Think about it. You spend more waking hours with your spouse than anyone else in your life. That relationship fundamentally shapes your daily mental health and long-term happiness. Yet most couples only seek help when things are already falling apart. In my time at Calm, I saw this pattern constantly. People would invest in meditation apps, personal therapy, wellness programs. But when it came to their relationship, literally the most important dynamic in their lives, they'd avoid getting help until it was almost too late. ### Why couples avoid therapy After talking to dozens of people about this, three barriers keep coming up: **Stigma.** There's this pervasive belief that needing help means you've failed. The framing is all wrong. Couples therapy shouldn't be viewed as emergency medicine. It should be relationship maintenance and growth. But the market hasn't successfully repositioned it that way. **Cost.** Traditional couples therapy runs $150-300 per session. Most people can't afford weekly sessions at that rate. So they either don't start, or they quit after a few visits before making real progress. **Accessibility.** Finding a good couples therapist is genuinely hard. Long wait lists. Limited availability. Geographic constraints if you're not in a major city. The whole system is designed for people who have unlimited time and money. The result? The market only serves two extremes: wealthy couples who can afford ongoing therapy, and couples in such severe crisis they have no choice. Everyone in between, which is most people, gets nothing. That's a massive market failure. And a huge opportunity. [Figure: The couples therapy market gap: the wealthy can afford traditional therapy and people in crisis eventually seek it, but the vast middle is priced out and underserved. That middle is the opportunity.] The market gap: the wealthy and the in-crisis get served. The vast, underserved middle is the opening. ## How AI changes the economics What makes this moment interesting is that AI is genuinely changing mental health delivery in ways we couldn't have imagined five years ago. AI companion and mental health apps have been downloaded 220 million times globally as of mid-2025 [TechCrunch](https://techcrunch.com/2025/08/12/ai-companion-apps-on-track-to-pull-in-120m-in-2025/). Millions of people are using tools like ChatGPT and Claude as personal therapists. Some of them say they feel closer to an AI companion than to the people actually in their lives, which is simultaneously fascinating and slightly terrifying. AI is also genuinely good at this particular job. It excels at pattern recognition. It can identify communication patterns, attachment styles, and recurring conflict triggers across thousands of data points. It never gets tired. It's always available. And it can provide support at a fraction of the cost of traditional couples therapy. Look, I'm not saying AI should replace human therapists. That's not the play here, especially in something as sensitive as relationships. But AI can dramatically reduce the cost and accessibility barriers that keep most couples from getting help in the first place. The play is AI augmenting human expertise, so that expertise reaches people who otherwise couldn't afford it or access it. ## What that model looks like in practice My client, Deeply, is building in this space, and their model is a useful picture of the shape. They're AI-enabled, but built so the human element that makes therapy work stays in. The product is a bilateral relationship assessment. Think of it like an MRI for your relationship. Both partners do structured interviews with trained professionals. AI analyzes the patterns across all those data points. Expert therapists review everything and produce a diagnostic report in about a week. What comes back isn't generic "communicate better" advice. It's specific scripts. Exact interventions tailored to that couple's dynamics. Root causes, not just symptoms. Concrete steps to start on now, foundations for the next 30 days, and a roadmap for the year. From there the couple can take the assessment into traditional therapy, work through a guided program, or use it as a map for their own work. ### Why the shape works After working in mental health at Calm and education at MasterClass, I've learned something crucial: the best products remove barriers while maintaining quality and trust. Three things about this shape do that. **Speed.** Traditional couples therapy takes months to identify root causes. An assessment like this gets there in days. For couples who are struggling, that speed matters enormously. For couples who are doing fine but want to strengthen things, speed is what makes it feasible to fit into a busy life at all. **Diagnosis, not treatment.** It isn't replacing therapy. It's telling a couple what's actually going on, which is the part traditional therapy spends months arriving at. A couple who walks into a therapist's office already knowing the pattern skips that discovery period entirely. **Accessibility.** At a fraction of the cost of traditional couples therapy, this becomes something people can actually afford. You're not choosing between therapy and other financial priorities anymore. [Figure: Traditional couples therapy takes months to deliver insight. AI-enabled assessment delivers it in days.] The economics change when insight arrives in days instead of months. ## The opportunity is the middle From a growth and product perspective, the part that gets me is how empty the field is. The category is stuck in 2005. Most solutions are just "find a therapist near you" directories. There's no Calm for couples. No Headspace equivalent. Nothing modern, accessible and scalable. The addressable market is enormous. Think about how many couples exist who would benefit from relationship support but aren't getting it: - Couples who are doing fine but want to be great - Couples who are drifting apart slowly but aren't in crisis yet - Couples navigating major life transitions (new baby, career changes, relocation) - Couples who tried traditional therapy but couldn't afford to continue - Couples who want help but the wait list is 3 months long All of these people are unserved today. The market caters to wealthy couples or couples in severe crisis, and to nobody else. That middle is reachable now, because AI takes the cost and the wait out of the equation. ## What this means for product people I'm not just excited about this as a growth advisor. I'm excited about what it says for how we build products in sensitive, high-trust categories generally. ### 1\. AI as co-pilot, not replacement This is the model that actually works when trust is the constraint. Pure AI solutions feel cold and risky when the subject is your relationship. Pure human solutions don't scale and stay expensive. The hybrid, with AI doing the pattern recognition and analysis and humans doing the insight delivery and trust-building, is the sweet spot. We're seeing this work in education (Khan Academy's AI tutor), coding (GitHub Copilot), healthcare diagnostics, and now mental health. The pattern is clear: augment human expertise, don't replace it. ### 2\. Removing stigma through product design One reason Calm succeeded was making meditation feel normal and accessible instead of something you only do if you're stressed out. The product design communicated: this is for everyone, this is self-investment. This category needs to do the same thing. The framing can't be "fix your broken relationship." It needs to be "see clearly, understand deeply, grow intentionally." That shift in positioning matters enormously for adoption. It should feel like going to the gym or getting an annual physical, not like admitting failure. ### 3\. Solve accessibility first Most mental health products focus on efficacy first, accessibility second. But if your solution is highly effective and nobody can access it, you haven't solved the real problem. The biggest issue isn't that therapy doesn't work. It's that most people can't get to it, because of cost, time, or availability. Solving accessibility first is the unlock, and it's what opens up the underserved middle. ### 4\. Meet couples where they are Not everyone needs or wants ongoing therapy. Some couples just want clarity on what's actually happening in their relationship. Some want a roadmap they can follow on their own. Some want to accelerate work they're already doing with a therapist. Serve all three instead of forcing everyone into one model. ## Why now The timing is right for a few reasons: **AI capabilities have crossed a threshold.** Five years ago, AI couldn't do this kind of nuanced pattern recognition well enough to be trusted with something this sensitive. Now it can. **Consumer comfort with AI has increased.** 220 million downloads of AI companion apps [TechCrunch](https://techcrunch.com/2025/08/12/ai-companion-apps-on-track-to-pull-in-120m-in-2025/) shows people are already comfortable using AI for emotional support. The stigma of "talking to a bot" has largely disappeared. **The market is ready to reframe couples therapy.** Younger generations are more open to therapy in general and more willing to invest in relationship health proactively. The stigma is cracking. **Economic pressure is real.** With traditional therapy getting more expensive and harder to reach, people are actively looking for affordable alternatives that actually work. All of these create a window for someone to build the modern version of this. The opportunity is sitting there, largely untapped. ## What I'd tell a founder building here A few lessons I'm taking from this engagement: **Don't underestimate trust signals.** In relationships and mental health, trust is everything. Having expert therapists review every piece of AI analysis is the reason Deeply's output gets acted on. Don't cut the human element to save costs; it's what makes people believe the answer enough to use it. **Identify your ICP clearly.** Deeply is focused on analytically-minded couples who want to understand the system, not just vent feelings. That specificity makes everything else easier: positioning, pricing, product development, marketing. **Speed to insight matters.** People don't want to spend months in therapy just to find out what's wrong. Compress the time to insight without giving up quality and you have a real advantage. ## The takeaway The market is underserved in the middle. What exists is expensive, hard to reach, and stigmatized enough that most people wait for a crisis before they touch it. AI changes that math by making expert insight cheap enough and fast enough to reach the people who were never going to get it otherwise. This isn't about replacing therapists. It's about removing the barriers that keep most couples from getting help at all, whether they're in crisis or just want to make a good relationship better. The opportunity is enormous, and almost nobody is building for the middle. ## Common questions ### Why is the couples therapy market considered broken? Because it only serves the two ends. The wealthy can afford traditional therapy, and people in crisis eventually seek it out, but the vast middle is priced out and held back by stigma, so most relationships get no support at all despite the spouse being the single biggest factor in a person's mental health. ### How does AI change couples therapy? It collapses the cost and time. AI pattern recognition paired with expert therapists can deliver specific relationship insights in days instead of months, at a fraction of traditional cost. It does not replace therapists; it makes real expertise accessible to people who could never afford it before. ### Is couples therapy only for relationships in crisis? No, and that framing is part of the problem. It is valuable whether you are thriving or struggling, six months in or sixteen years in. Treating it as a last resort is exactly the stigma that keeps the underserved middle from getting help. --- ## The 4 Stages Of A Good Growth Process Source: https://ottergrowth.com/notes/the-4-stages-of-a-good-growth-process Published: 2020-07-26 > Growth is not a bag of silver-bullet ideas. It is a repeatable process with four stages: user-centric ideation, disciplined prioritization, high-velocity execution, and focused learnings. Each has a target outcome, a specific way it breaks, and a self-assessment checklist below. Run the cycle one stage at a time, and manage against the failure points rather than the target outcomes, because the failure point is what is actually happening on your team. Ask a growth lead which stage they are weakest at and it is almost always the last one, which is the one nobody has ever made them defend. When I evaluate growth candidates for hire, I test specifically on these 4 areas of a growth process. Not their favorite growth hack, and not their channel expertise. I ask them to walk me through their process, and then I ask which of the four stages they are weakest at. The answer to that question is at the bottom of this piece, because it is the same answer almost every time. Each stage below has a target outcome, a specific way it breaks, and a checklist. **The failure points are the part to pay attention to.** A target outcome tells you what good looks like, which everyone already agrees on. The failure point tells you what is actually happening on your team right now, and that is the thing you manage against. Why? "Growth" is a process. The best growth leaders I know are successful because they run a diligent and repeatable process which drives business learnings and ultimately revenue growth metrics. There are no "silver-bullet" ideas in a good growth program, there is only user-centric ideation, disciplined prioritization, high-velocity execution, and focused learnings. From startups all the way up to established companies, understanding these concepts is what separates winners from losers. It's what enables consistent outperformance in a market. ## The 4 stages of a growth process [Figure: The four stages of a good growth process run in sequence: user-centric ideation, then disciplined prioritization, then high-velocity execution, then focused learnings, which feeds the next round of ideation.] Growth is a repeatable process, not a bag of silver bullets. The four stages are circular: the learnings feed the next round of ideation. I'll go into each of these stages in more detail including target outcomes, common failure points, and a self assessment checklist. But first, some definitions - **User-Centric Ideation:** This is the process for coming up with hypotheses in a user-centric way. Meaning that you root your hypotheses in foundational user understanding and research rather than personal or company-held opinions/perspectives. **Disciplined Prioritization:** A framework for capturing and evaluating the ROI of hypotheses in an unbiased way. And a backlog that is resistant to turbulence and new ideas/requests which are low-ROI. **High-Velocity Execution:** This is referencing your experiment velocity which is the top-of-funnel for a good experimentation program. How many experiments do you launch per week, per person on your team? **Focused Learnings:** How to get the most learnings benefit from positive and negative results, sharing learnings amongst your team and broader org to amplify collective intelligence, and grouping learning goals into "themes". This helps guide you towards a comprehensive understanding of users. * * * The process of growth experimentation and iteration is circular. Which means, if you are not strong in every stage, then the strength of the whole cycle suffers. Weak learnings produce weak ideation. Weak ideation produces a weak backlog, and no prioritization framework saves that. So if you are trying to improve a growth metric, there is nothing more important and impactful than also improving in these areas. * * * ## User-centric ideation [Figure: Stage 1 of 4: user-centric ideation.] Generate a large, high-confidence backlog from many idea sources, not one. **Target outcome:** - A large backlog that is high-confidence and derived from multiple idea source types. **Most common failure point:** - One low-confidence source, usually a brainstorm, and you call it a backlog. **Self-assessment checklist:** **Have you...** - Drawn from multiple idea sources? I.e. Comparables, user research, data analysis, group brainstorming, subject matter experts etc. - Generated at least 10 ideas? - Structured the ideas around a common user theme/question or metric you are trying to solve for? * * * ## Disciplined prioritization [Figure: Stage 2 of 4: disciplined prioritization.] Force-rank the backlog by ROI and risk, resistant to the loudest voice in the room. **Target outcome:** - A sort-ranking of your ideas which optimizes for return-on-investment and the lowest risk. **Most common failure point:** - Effort and confidence get skipped, and you run whatever the loudest voice in the room wants next. **Self-assessment checklist:** **Have you...** - Used a quantifiable framework to estimate the ROI ideas such as [I.C.E. Scoring](/notes/impact-confidence-effort-i-c-e-scoring)? - Evaluated the confidence of an idea accurately by taking into account the idea source type? - Proactively communicated your prioritization method to stakeholders and ran new ideas or requests through the same process as they come up? - Gotten at least one other source of scoring input to mitigate your personal bias and strengthen scoring confidence? - Applied the framework on multiple levels to maximize your team's effectiveness? I.e. Initiative/user-theme level, and individual hypothesis level. * * * ## High-velocity execution [Figure: Stage 3 of 4: high-velocity execution.] A streamlined process that ships experiments fast, measured as experiments per week per person. **Target outcome:** - A high experimentation velocity (rate of experiments / week / team size) **Most common failure points:** - No real process at all. The ideas exist and nothing ships. **Self-assessment checklist:** **Have you...** - Created a streamlined process which is straightforward, reduces back-and-forth + idle time, leverages the strengths of different functions on your team, and empowers individuals with ownership? - Set up, automated, and consolidated the tools you use for execution? Adequately trained your team on how to use them and set reasonable expectations? - Documented and gotten feedback on your processes? - Designed experiments which are inherently fast because they are focused on testing the core underlying hypothesis and have de-scoped features which are not crucial to that goal? - Developed a growth-mindset and culture on the team which prioritizes experiment execution speed over comprehensiveness? * * * ## Focused learnings [Figure: Stage 4 of 4: focused learnings.] Extract and share the signal from every result, positive or negative, to compound team intelligence. **Target outcome:** - A better understanding of your users to help improve strategic direction and generate higher quality hypotheses across your team and the broader org. **Most common failure point:** - You ship the test, read the result alone and move on. Nobody else gets smarter. Dumb way to spend a test budget. **Self-assessment checklist:** **Have you...** - Created a standardized results analysis report dashboard with multiple segmentation views which allows you to understand the drivers of positive and negative results in an experiment? - Ensured that you are executing statistical best practices such as a minimum runtime duration and minimum statistical significance? - Developed a way to document experiment results that is easy to understand and digest for others? - Created effective sharing mediums for your team and broader org? - Proactively shared both positive and negative results? - Tied results back to an overarching user question that you are trying to figure out? * * * ## The answer to the interview question It is the last one. Almost always. Not because learnings is the hardest stage. Because it is the one nobody has ever made them defend. Which matters more than it sounds, because the cycle is circular. Weak learnings produce weak ideation. Weak ideation produces a weak backlog, and no prioritization framework saves that. Be weak at one stage and the whole cycle is weak, and the stage nobody defends is the one quietly setting the ceiling on the other three. ## Common questions ### What are the four stages of a good growth process? User-centric ideation, disciplined prioritization, high-velocity execution, and focused learnings. The process is circular rather than a sequence that ends: better learnings produce better ideas, which is what makes it compound. Be weak at one stage and the whole cycle is weak. ### Where do most growth programs fail? Almost always at one specific stage. They ideate from a single low-confidence source like group brainstorming, they prioritize without honestly weighing effort and confidence, they never build a real execution process so velocity stays low, or they run experiments and never share the learnings. Find your weakest stage and fix that one first. ### How do I know if my growth process is actually working? Pressure-test each stage against its target outcome. Ideation: a large backlog drawn from multiple idea sources. Prioritization: a ROI-ranked backlog that survives new requests. Execution: a rising rate of experiments per week per person. Learnings: a standard results format that both positive and negative outcomes get shared in. If any stage can't clear its bar, that's your bottleneck. --- ## ICE scoring: how to prioritize a growth roadmap on impact, confidence and effort Source: https://ottergrowth.com/notes/impact-confidence-effort-i-c-e-scoring Published: 2020-02-24 > ICE force-ranks a growth backlog so the loudest voice in the room doesn't set the roadmap. Score each idea on impact (relative, 1-5), effort (person-weeks), and confidence (50/75/90%, based on the source), then rank by (impact ÷ effort) × confidence. Double down on the top quarter, take quick wins from the next, and skip the bottom half this quarter. Confidence is the factor almost everyone ignores, and it's the one that matters most. Within a growth lever that we are focusing on, how do we determine what initiatives to focus on and which to not pay attention to? One of the easiest ways is to apply a simple scoring method in order to force-rank your roadmap. ICE stands for impact, confidence, and effort. [Figure: ICE is a ratio of impact over effort, wrapped in confidence. Impact sits over effort inside an inner circle, and confidence surrounds the whole ratio as an outer ring.] ICE optimizes impact per unit of effort, with confidence surrounding the whole ratio: low confidence means the impact could be lower, or the effort higher, than you think. It is intentionally designed to be a ratio of Impact over Effort, surrounded by confidence. This is because you want to be optimizing for maximum impact to your target metric, per amount of effort spent. Confidence underscores Impact and Confidence because with low confidence, the impact could be much lower than expected or the effort could be much higher than expected. ## How to put I.C.E. into practice In many cases, people only think of the impact, and not the required effort, and in even more cases, people don’t consider confidence at all. **Impact** is the amount we believe we can improve the target metric. - We grade impact on a simple 1-5 scale. What is important is that we are grading relative to other items on the list, since we don’t know what the actual impact will be, but that isn’t important for a relative-scoring exercise.  **Effort** is the number of weeks of a person’s time. - We grade up by half-weeks with no maximum. Effort estimation isn’t intended to be exact, but again, for relative-ranking, it is only important to be somewhat accurate. **Confidence** is derived from the source of the information. - Confidence can have a score of 50%, 75% or 90% (we're never 100% confident). A 90% means that we have hard data, strong comparables, or direct expertise which backs up our belief in the impact score. A 75% means that we have indirectly related data, or evidence to support our scoring. 50% means that we could be right or wrong, we don't have significant data to support or disprove our scoring. ## Confidence truly is key Confidence is the most overlooked of the factors, but is arguably the most important. An easy way to understand confidence is to think about information sources for things like scientific discoveries, news publications and others – your confidence in that information relies on the credibility of the source and how well it is corroborated.  There are a number of idea sources of information, with differing levels of confidence. [Figure: Idea sources ranked by confidence, where confidence equals data plus relevancy. From highest to lowest: prior experiments, data and analytics, comparable products, user research, subject-matter experts, and brainstorming, which is the lowest-confidence source.] Confidence is a mix of data and relevancy. Brainstorming, the source most teams lean on, sits at the bottom. Most people are familiar with brainstorming, which is actually the lowest confidence idea source. The equation that we consider is that confidence is made up of a combination of data and relevancy. We at MasterClass ensure that we have a diverse backlog from a number of different information sources with high confidence. [Figure: Three high-confidence idea sources practiced at MasterClass: teardowns of sixty comparable companies, subject-matter-expert surveys with free product access, and on-site user polls.] A diverse, high-confidence backlog at MasterClass drew from many sources at once. Three of the most practiced: - We do teardowns of 60 comparable companies cross Entertainment, Education, and Consumer Tech verticals.  - We survey subject matter experts and give them free access to the product. - We do on-site polls and ask users directly what feature in the backlog would be most beneficial towards their purchase decision. These are just a few of our most practiced idea sources, but in reality we do them all. ## Using I.C.E. to drive focus across initiatives [Figure: The ICE formula: impact divided by effort, in parentheses, multiplied by confidence, equals the value score.] The value score is the ratio of impact over effort, factored by confidence. And so, the equation is the ratio of impact over effort, factored by confidence. ## What the third term actually does to two ideas Take two items off the same backlog and run them through it. | Idea | Impact | Effort | Confidence | Score | | --- | --- | --- | --- | --- | | Paywall copy test, off a strong comparable | 3 | 0.5 wks | 90% | **5.4** | | Rebuild onboarding, off a brainstorm | 5 | 6 wks | 50% | **0.42** | The onboarding rebuild is the one the team wants to build. It scores highest on impact, it is the ambitious one, and it is the one that gets talked about in the room. It also ranks thirteen times below a copy test. That is the whole argument for the third term. A shaky idea is not just uncertain. Its impact is probably overstated and its effort is probably understated, and both of those errors push it up the list. Drop confidence from the formula and you have not built a prioritization framework, you have built a very tidy machine for ranking your own guesses. We can then use the value score to sort the backlog from high --> low score. 1. Then we take the top ¼ of the backlog and those are things we double down on. 2. The next ¼ we continue to evaluate and invest in quick wins. 3. The bottom half are things that we determine that we aren’t going to focus on for the quarter. [Figure: After sorting the backlog by value score: double down on the top quarter, take quick wins from the next quarter, and do not work the bottom half this quarter.] Sort the backlog high to low by value score, then split it: double down, quick wins, or don't do it this quarter. The techniques we've discussed are fairly simple, but have a huge impact on the strength of a backlog - 1.) Applying an impact/effort ratio, 2.) Leveraging idea sources that have high confidence, and 3.) Using a value score to determine what you should invest in, continue evaluating, and not do. In the next post in this series I'll discuss core growth levers to evaluate and the basic structure of a Growth Model! ## Common questions ### What does ICE stand for? Impact, Confidence, and Effort. The score is a ratio: (impact ÷ effort) × confidence. Impact and effort are scored relative to the other items on your list, and confidence scales the whole thing based on how much you trust the estimate. ### Why is confidence the most important factor? Because low confidence means your impact estimate could be far too high, or your effort far too low, and either one breaks the ranking. Confidence comes from the source of the idea: hard data, strong comparables, or direct expertise is about 90%; indirectly related evidence is about 75%; a hunch with little to back it is 50%. Most people skip this factor entirely, which is exactly why their roadmaps drift. ### How do I turn ICE scores into a roadmap? Sort the backlog high to low by value score, then split it. Double down on the top quarter, take the quick wins from the next quarter, and don't work the bottom half this quarter. Re-run the same scoring on any new idea or request so it competes on the same terms as everything already in the backlog. --- ## How To Use And Optimize A Growth Model Source: https://ottergrowth.com/notes/how-to-use-and-optimize-a-growth-model Published: 2020-07-19 > A growth model does for resourcing what a financial model does for budgeting: one sheet showing how each lever feeds revenue, CAC and LTV, so you can see what moves what before you commit the quarter. There is a worked one-month model below that you can copy. Structure it around the sub-metrics you can actually affect, and you can see what a 10% change in one metric does to company growth, and trade efficiency for scale on purpose instead of by accident. I'd argue that the two most important decisions you can make as a leader are 1) who to hire and 2) where to allocate resources. Having a growth model is key to number 2. It's interesting how we put significantly less time and energy into our resource allocation decisions relative to our hiring decisions. This is because, well, it's not easy to measure and evaluate the resourcing tradeoffs across an organization. But it can be, using a growth model. * * * ## Issues with resourcing decisions today: One of the most important questions companies should be asking themselves regularly - **What is the most effective allocation of resources to maximize our company's growth?** Most companies determine this every quarter by doing some version of an "executive round-table" debate supplemented by business cases. That approach has a couple important issues with it that are worth considering given how crucial these decisions are: 1. It is subject to "loudest voice in the room" risks where decisions are influenced by how strong of a pitch someone gives or how much weight in the company a person has. This approach is open to bias and opinions. 2. It is exacerbated by the fact that the business cases fail to provide a unified way to understand, quantifiably, the tradeoffs between resourcing different areas of the business. They are often created by department heads and subject to bias as it is. And oftentimes in the business cases, estimated impact isn't factored by the amount of resources needed either, so this makes apples to apples comparisons impossible. **A growth model is the answer to those issues.** ## What that actually looks like when it goes wrong A third of a quarter's roadmap going to work nobody had ranked. Not bad ideas. Three teams each getting a little of what they wanted. I watched this at a marketplace recently. Engineering capacity was tight, every team lead had something they needed staffed, and nobody wanted the conflict of saying no outright. So the quarter got sliced. A bit here, a bit there, everyone partially satisfied. Losing the roadmap space is the cheap part. The expensive part is momentum, because that work runs on different cadences, needs constant cross-functional alignment, and drags the velocity of everything around it. And the actual problem there was not prioritization. Every team had its own metrics and goals, nothing said which metric outranked which, and nothing said whether one team's goal had been approved for another team to spend its engineers on. When that is unclear, the strongest pitch wins instead of the strongest case. That is a popularity contest with a spreadsheet attached. * * * ## Defining a "Growth Model" A growth model is similar to a financial model, except it is structured around growth levers and metrics which you have the ability to affect. With how important resourcing decisions are, it is kind of crazy that many companies don't put in the time and diligence to create one in order to understand the relationship across growth levers. Here are some of the most common growth levers. This differs business to business and B2C vs. B2B: [Figure: Common growth levers: paid, content, viral, and partnerships on the acquisition side, plus monetization, retention, and new markets. Each differs by business and by B2C versus B2B.] Common growth levers. The exact set differs business to business and B2C versus B2B. How do you know what will be the best use of your company’s time and resources? Is there not a more important decision to be made than where you put your resources? * * * ## Structuring a growth model A growth model, structured around your growth levers and metrics which you can affect, will help you understand the relationships and tradeoffs between them. It will also give you a shared framework and language to use and evaluate these tradeoffs. Here is an example of what a growth model structure could look like at a high-level: [Figure: A growth model structure: growth levers feed the sub-metrics you can affect, which roll up into company metrics: CAC, LTV, and revenue.] Each lever has metrics you can move. Those roll up into company metrics that span every lever: CAC, LTV, and revenue. A growth model helps zero-in on which sub-metrics you are planning to affect and how that translates to revenue for the company. They can also be used to validate assumptions and stress test baseline and upside scenarios. The metrics within each growth lever are key metrics which you have the ability to affect. The output metrics, or company growth metrics, have a volume metric (i.e. # acquired) and a health metric (revenue / # of visitors). Also, you will notice that there truly is a relation across all of these growth levers - Revenue, and Customer Acquisition Cost / Lifetime Value. * * * ## A worked example: one month on one sheet Here is the simplest version of the whole thing, deliberately generic, in the shape a real one takes. One month, four blocks, and every number below the first block is derived rather than typed in. **Block 1. The money and what it buys.** | Input | This month | | --- | --- | | Paid spend | $200,000 | | Cost per install | $4.00 | | Installs | 50,000 | Spend and cost per install are the two you control directly. Everything below here is a rate, not a decision, and that distinction is most of the value of laying it out this way. **Block 2. The rates you can move.** | Rate | Out | | --- | --- | | Install to trial, 12% | 6,000 | | Trial to paid, 35% | 2,100 | This block is product work, not budget. **Block 3. What a customer is worth.** | Input | Value | | --- | --- | | Revenue per payer | $12 / mo | | Monthly churn, 8% | 12.5 months | | Lifetime value | $150 | Churn is doing the work here, not price. At 8% a month the average life is 12.5 months. Halve the churn and the lifetime doubles. **Block 4. What comes out.** CAC $95. LTV $150. **LTV:CAC 1.6.** Payback 7.9 months. One number is what the whole sheet exists to produce, and 1.6 is thin. Now change one input. Take trial to paid from 35% to 45% and you get 2,700 payers instead of 2,100. CAC falls to $74, LTV:CAC goes to 2.0, and you spent nothing extra. That is the answer the sheet gives you and the round-table does not: the cheapest route from 1.6 to 2.0 runs through block 2, not through more spend. You cannot see that by arguing about it, and you cannot see it from a business case that estimated impact without dividing by the resources needed. * * * ## The output of a good growth model With a growth model, you are able to understand what a 10% increase in a sub-metric means for Revenue and CAC/LTV. [Figure: A growth model separates health from growth. Health: each dollar less spent or more earned per acquired improves CAC and LTV. Growth: each additional visitor, acquired, or retained user adds revenue.] The model makes the health-versus-growth tradeoff explicit. You can trade efficiency for scale on purpose, instead of assuming you can maximize both at once. So, when you have a growth model that calculates changes in one metric in one growth lever of your business, you will be able to see the overall impact to your **company growth** level metrics, which span across growth levers and allow you to make these decisions. One thing I've pointed out in the diagram above is the concept of health and growth metrics which is something that should be called out throughout a growth model. You can trade health for growth in many cases depending on your business priorities. The clearest example of this is paid marketing - where you can trade efficiency for scale. However, if you don't have a growth model in place and concept of health vs. growth metrics, an unrealistic expectation can arise around achieving both simultaneously. Another benefit of a growth model is that it forces you to talk about which sub-metrics you believe you can change within a growth lever and what your hypotheses are to change them, whereas, with business cases this level of detail is not typically achieved, and there are sweeping assumptions made to guestimate what the net revenue impact will be of an initiative. **But perhaps the most underrated benefit** of a growth model is that it provides a **shared language** and **basis for discussion** when it comes to growth planning and resourcing which typically tends to be segregated by department rather than unified by a shared company growth model. * * * ## What to put around it The model is the instrument. It still needs something to point at, or you are back in the round-table with better arithmetic. 1. **Three to five Tier 1 metrics for the whole company.** Not per team. Company. 2. **Put them in priority order.** Everyone skips this, and it is the part that makes a trade-off conversation possible when two of them pull against each other. 3. **Decide in quarterly planning which teams support which metrics.** Explicitly, on the record. 4. **Fund secondary metrics out of slack time,** roughly 10 to 15%, rather than out of the committed roadmap. Then every incoming request answers four questions. What metric does this move? Is that metric Tier 1? Has the exec team already aligned on it? And do we want this team carrying it this quarter? A growth model is what makes the first of those answerable with a number instead of a pitch. ## Common questions ### What is a growth model? A model, structured like a financial model, but built around the growth levers and sub-metrics you can actually affect. It shows how a change in one metric, say a 10% lift in conversion, flows through to company-level outcomes: revenue, CAC, and LTV. ### Why use a growth model instead of business cases? Because business cases give you no way to compare resourcing tradeoffs apples-to-apples. They're built department by department, biased toward whoever writes them, and usually don't factor impact by the resources required. A growth model puts every lever in one quantified frame, so the decision turns on the numbers instead of the loudest voice in the room. ### What's the difference between health and growth metrics? Health metrics measure efficiency: revenue per visitor, CAC, LTV. Growth (or volume) metrics measure scale: visitors, users acquired, users retained. You can often trade one for the other, paid marketing being the clearest example. A growth model makes that tradeoff explicit so nobody assumes you can maximize both at the same time. --- ## The paywall audit: three levers, and the five things people get wrong Source: https://ottergrowth.com/notes/the-paywall-audit Published: 2026-09-09 > The paywall is one of your highest-leverage surfaces for LTV:CAC, and it runs on three levers rather than one: how well it converts, where it is placed and exposed after value has landed, and the LTV of the plan people end up on. Win one and lose the other two and you get a paywall that looks like it is working. Below are the five things I see missed most, in the order they cost you, each with its publisher and its caveat. The paywall setup with your best day-one conversion can be the one with your worst twelve-month LTV. Adapty has found exactly that: the configuration with the lowest Day 0 conversion often produces the highest twelve-month value. Which is why "how is our paywall converting" is a third of the right question. ## The three levers **One: how well it converts.** The screen itself. Framing, defaults, what it promises, what it does with an objection. **Two: where it is placed, and how it is exposed after value has landed.** A paywall is not just a screen, it is a moment. The same screen at a different moment is a different product. **Three: the LTV of the plan people end up on.** Two paywalls can convert identically and be worth very different amounts, because one of them lands people on a plan they keep and the other lands them on a plan they cancel. Almost every paywall conversation I get pulled into is about lever one. Most of the money is in levers two and three, and lever three is the one nobody has looked at. ## The five things people get wrong ### 1. The paywall arrives after intent has already decayed 89.4% of all trial starts happen on Day 0, and 82% of purchases happen in the first session (Adapty 2026). Onboarding-placed paywalls with trials convert at 1.78% against 0.89% for in-app placement. Rootd reported a 5x revenue increase after moving theirs to the start of onboarding, and a plant-care app went from 3% to 15% on sign-up-to-trial. Download intent decays fast, and users who delay a trial rarely come back to start one. That is lever two, and it is usually the cheapest of the three to move because it requires no new design at all. **The exception is real and it is Education.** Only 71.3% of Education trials start on Day 0, and 23.5% start on Day 31 or later. People in that category genuinely browse, leave and come back with intent, so hard-gating at the end of onboarding cuts off the return trip. ### 2. Annual is priced as a lump sum Re-anchoring the annual price to a monthly equivalent lifted new revenue per paywall impression by 45% in Brazil, 26% in Mexico and 8% in the USA, at zero price change (Mojo, first-party, reported via RevenueCat). Per-day pricing display lifts annual adoption by 20 to 40%, though that figure comes from industry reports rather than a single controlled study, so read it as directional. Look at the spread across those three markets, because it tells you what drives the effect: it is biggest where a lump sum feels prohibitive relative to income. ### 3. The default plan is wrong, or there is no default at all Switching the default from monthly to annual inverts the plan mix, roughly 37% yearly to 63% yearly. Among users who did not close the paywall, 82% simply took whatever was pre-selected. One app pre-selecting annual saw a 70% increase in annual revenue. That is Superwall network data, observational rather than randomized, so treat it as a strong prior rather than a proven causal effect. This is lever three hiding in plain sight, and it is the element nobody argued about. Almost every element on that screen was decided deliberately. The price. The headline. The badge, the artwork, the order of the benefit list. Someone argued about each one, probably in a meeting with a designer present. And then there is the toggle, which loads on whatever it loaded on the day it was built. [Figure: Share of paywall purchasers who take the pre-selected plan] Bar widths are the stated split. The pre-selected option is not a starting point, it is the answer for most buyers. That toggle is doing more work than the artwork. A lot of paywall performance is decided by the options you never framed as decisions, which is why redesigns so often move nothing: they change everything that was already argued about and leave the one thing that was not. ### 4. It lists features when it should promise an outcome Features answer "what do I get." Outcomes answer "what happens to me." Only one of those is why anybody downloaded. One meditation app replaced a features carousel with a preview of the first week's program and moved paywall conversion from 20% to 30%, a 50% lift. That figure comes from a secondary source, so treat it as medium confidence. The principle underneath it is well established and is not in doubt. ### 5. Nothing catches the people who say no, and nothing reassures the people who are scared of the trial The single biggest objection at a trial paywall is not price. It is the fear of forgetting to cancel. Blinkist's transparent trial timeline, showing the start day, the reminder day and the charge day, produced a 23% conversion lift, 55% fewer complaints, and push opt-in going from 6% to 74%. That one was A/B tested and reported first-party by their Head of Product, which makes it one of the better-evidenced numbers in this whole area. Then there is the other end. Exit offers contribute 15 to 20% of total revenue (Superwall), post-paywall welcome offers produce a 10 to 15% ARPU uplift (Adapty), and roughly 50% of users who start an in-app purchase never complete it. That last number is the one worth sitting with: half the people who got as far as tapping buy are still leaving, and most paywalls have nothing at all built for them. ## "Always offer a trial" is wrong for about half of app categories This belongs to lever three and almost nobody checks it. Adapty's category analysis has trials actively hurting LTV in four of eight categories: Productivity at -13.7%, Lifestyle at -21.2%, plus Entertainment and Graphics & Design. Utilities sees the largest premium at +85.1% and Health & Fitness +63.6%. Our own paywall library calls this the single most important nuance in the paywall benchmarking space, and it is the clearest example of why lever three exists as a separate lever. A trial can raise your conversion rate and lower what a customer is worth. Both things are true at once, and only one of them shows up in the dashboard people look at. ## The four questions So when someone asks me to look at their paywall, I do not start with the design. I ask four things, and I ask them in this order because the order follows how a user actually meets the screen. [Figure: Where the four paywall audit questions land on the screen] Three of the four questions are about things already on the screen. The fourth is about what is not. ### What does the toggle load on, and who decided that? The answer is almost always "it has always been that way." That is a fossil sitting on your highest-intent screen. It was set by whoever built the first version, usually before the pricing was final, and it has survived every redesign since because a pre-selected state does not look like a decision when you are reviewing a mockup. Ask who chose it. If nobody can name a person or a reason, you have found a test. ### Is the annual price doing the mental math for the user, or asking them to? A yearly figure presented as a lump sum asks the buyer to divide, at the exact moment they are least willing to do arithmetic. Most of them will not do it, and the ones who do are doing it while deciding whether to trust you. The monthly-equivalent framing costs nothing to implement and changes nothing about what you collect. ### Can they see when they will be charged? The fear at the decision point is not "is this worth it." It is "will I forget to cancel." If the screen does not answer that question, the user answers it themselves, and the safe answer is to leave. Showing the start date, the reminder, and the billing date turns an open-ended commitment into a bounded one. That is why making cancellation legible tends to make starting easier rather than harder. ### What happens on the screen right before this one? Half of what is wrong with a paywall is not on the paywall. If the user arrives without having felt the product do anything, the screen is being asked to sell a promise instead of confirming an experience. No amount of framing fixes that, because the problem happened before the paywall loaded. This is the question that most often redirects the whole engagement, and it is the one people are most surprised I ask. ## Why defaults survive one redesign after another Usually at least two of those four have never been examined. That is not sloppiness. Defaults do not announce themselves as choices. When you review a paywall mockup, you see a screen with a plan already selected, and your eye goes to the things that vary between versions: the headline, the artwork, the button copy. The selected state reads as part of the furniture, not as a variable, so it does not enter the conversation and it survives. The same thing protects the missing billing date. Nobody reviewing a screen notices the sentence that is not on it. ## The rule I would hand you Anything on that screen you cannot give a reason for is a test you have not run yet. Walk the paywall element by element and try to say out loud why each one is the way it is. The elements with real answers are decisions, and decisions can be defended or revisited on purpose. The ones that produce a shrug are your backlog, ranked by how much traffic passes through them. The toggle is usually at the top of that list, because it has the most traffic and the least deliberation of anything on the screen. ## What a paywall audit actually hands you Since this is what I do, here is the shape of it rather than a pitch. The body is short on purpose: our goal, a two-to-three sentence read, a rating stated against the achievable ceiling of the fixes rather than against any observed conversion number, "do these now" grouped by paywall element (plan cards, billing toggle, trial CTA), and the numbers that matter with one number per metric. **When no funnel data was supplied, that section becomes "here are the bars" and says so explicitly rather than implying a measurement nobody took.** The appendix carries the benchmark table in full form with its spreads and read-as caveats intact, what we observed, the full ICE-ranked backlog, and paywall comparables when a competitor set exists. Minimal input is paywall screenshots plus price points, or a public pricing URL. Without screenshots or a price the run does not start, because there is no version of this built from a description of a paywall instead of the paywall. Run against a public pricing page with zero funnel data, the top five moves in one sample were all framing: express the annual price monthly, restate the saving in real money, flip the default toggle, split the plan grid by who is buying, and put a trial timeline under the CTA. Not one of them touched the price. That was a worked sample on a public company rather than a client engagement, which is the honest way to describe it. ## The takeaway Three levers, and most paywall work only touches one. How well it converts, where it sits relative to the moment value lands, and the LTV of the plan people end up on. Before you brief a redesign, run the four questions: what the toggle loads on and who chose it, whether the annual price is doing the arithmetic or outsourcing it, whether the billing date is visible, and what the user experienced on the screen immediately before. Most teams find at least two answers they cannot defend, and those two are cheaper to change than anything a redesign would touch. ## Common questions ### What are the three levers on a paywall? How well it converts, where it is placed and exposed relative to the moment value lands, and the LTV of the plan people end up on. Most paywall work only touches the first. Two paywalls can convert identically and be worth very different amounts, because one lands people on a plan they keep and the other on a plan they cancel. ### Should every subscription app offer a free trial? No, and this is the least-checked decision in the category. Adapty's category analysis has trials hurting LTV in four of eight categories, Productivity at -13.7% and Lifestyle at -21.2%, against Utilities at +85.1% and Health & Fitness at +63.6%. A trial can raise your conversion rate and lower what a customer is worth at the same time. ### Why is my paywall not converting even after a redesign? Because redesigns change the elements that were already deliberate and leave the ones that were not. The plan toggle's default state, the absence of a visible billing date, and the framing of the annual price are rarely revisited in a design review, since none of them read as a variable when you are comparing mockups. Superwall's data has 82% of non-closing users taking whatever plan was pre-selected, so the default is carrying more conversion weight than the visual design it survived. ### What should I check on my paywall before redesigning it? Four things, in order: what the plan toggle loads on and who decided that, whether the annual price is expressed as a monthly equivalent or as a lump sum, whether the user can see the date they will be charged, and what the screen immediately before the paywall did for them. Each is cheap to change and none requires new design work. ### Does showing the billing date on a paywall hurt conversion? It tends to help. The hesitation at the purchase decision is usually about forgetting to cancel rather than about the value of the product, so a visible start date, reminder date, and billing date removes the specific unknown that causes people to postpone. Products that add the timeline generally see both conversion rise and billing complaints fall. --- ## Paywall optimization: six changes that lift conversion without touching price Source: https://ottergrowth.com/notes/paywall-defaults-and-framing Published: 2026-09-02 > Most paywall advice is discount advice wearing a costume. A discount buys conversion by giving away margin, permanently, for everyone who would have paid full price anyway. A default or a reframe buys it for free. Six that never touch price: re-express annual as a monthly equivalent, default the toggle to annual, put the trial timeline on the screen, move the paywall to the end of onboarding, pick trial length deliberately, and add a higher-priced plan instead of cutting the lower one. Run the framing and the default first: they are the cheapest to ship and they move the plan mix, which decides everything downstream. When conversion is soft, the first suggestion in the room is almost always a discount. Sometimes it is dressed up as a promotional tier or a limited-time offer, but it is the same move: buy conversion by giving away margin. It works, briefly. The problem with that move is that it never comes back. You have reset what your product is worth for everyone who sees the offer, including the people who would have paid full price. Every change below moves conversion without touching what you charge. Each one has a named source and a real effect size behind it. ## Re-express annual as a monthly equivalent Show "$4.99/month, billed annually" instead of "$59.99/year." Same price. Same billing. The only thing that changed is the unit the user does the mental math in, and the lump sum stops being the first thing they react to. Mojo reported new revenue per paywall impression up 45% in Brazil, 26% in Mexico, and 8% in the US on this change alone. Zero price change anywhere. Look at the spread across those three markets, because it tells you what is actually driving the effect. The effect is biggest where a lump sum feels prohibitive relative to income. If most of your growth is in markets where sixty dollars is a real decision rather than a rounding error, this is the highest-value item on the list for you specifically. ## Default the plan toggle to annual Whatever loads pre-selected reads as your recommendation, whether or not you meant it that way. Superwall's network data has switching that default flipping the plan mix from roughly 37% yearly to 63% yearly. Among people who did not close the paywall, 82% simply took whatever was already selected. [Figure: Plan mix before and after switching the paywall toggle default to annual] Segment heights are the stated percentages. The mix inverts on a change nobody sees. The second figure is the important one. Most of the people buying from your paywall are not comparing plans. They are accepting the one you put in front of them, which means the toggle is doing more work than the artwork, the copy, and the badge combined. ## Put the trial timeline on the screen Show the day the trial starts, the day the reminder goes out, and the day billing begins. Blinkist reported conversion up 23% and complaints down 55% after adding it. The first time I saw that result I assumed it was a fluke, because it runs against instinct. You are drawing attention to the charge. But it replicates. The reason is that the fear at the decision point was never "is this worth it." It is "will I forget to cancel." Showing the dates answers that question up front, so people start more readily rather than less. The complaints number is the proof: those were people who felt ambushed, and the ones who did not churn angrily were the ones who could see the date coming. ## Move the paywall to the end of onboarding Onboarding paywalls with trials convert at 1.78% against 0.89% for in-app paywalls, per Adapty's 2026 data. Rootd saw a 5x revenue increase after moving theirs earlier. [Figure: Onboarding paywall versus in-app paywall conversion rate] Bar lengths are the stated rates. Same paywall, different moment. The reason sits in a third number: Adapty has 89.4% of all trial starts happening on Day 0. Almost the entire decision happens in the first session, so a paywall placed later is competing for an audience that has already mostly left. The exception is education, where people genuinely browse, leave, and come back with intent. Hard-gating that behaviour at the end of onboarding cuts off the return trip. ## Pick trial length on purpose Long trials of 17 to 32 days convert at 42.5% against 25.5% for trials under 4 days, per RevenueCat's 2026 data. That looks like a settled argument until you add the other half. Short trials pay back ad spend faster, and cash conversion cycle is a real constraint when paid acquisition is your growth engine. Which is why RevenueCat has 46.5% of apps now running trials of 4 days or less, with the conversion rate that implies, on purpose. Both are defensible. A long trial buys conversion rate, a short one buys velocity, and which you need depends on whether your bottleneck is monetization or working capital. The other input is your category, because the right length is however long your value takes to arrive. Fitness apps often go about 30 days, since that mirrors a monthly habit cycle and the habit is the thing being sold. Education and wellness tend to land at 7 to 14, because intent drops fast and a month is long enough for the reason someone signed up to stop feeling urgent. Picking one by accident is the failure. And most trial lengths I look at were set once, early, by someone copying a competitor, and have never been revisited against the economics they now sit inside. ## Add a higher-priced plan instead of cutting the lower one When conversion is soft, the instinct is to add something cheaper. Try adding something more expensive. Superwall found that adding an annual plan alongside monthly drove proceeds per user up 80%. A plan almost nobody buys still does work, because it gives the target plan something to be measured against. Without a reference point, a price is just a number the user has to evaluate in a vacuum, and in a vacuum every price feels like a lot. This one is additive rather than a change. You are adding an option, not editing one, so nothing about your existing offer moves and nothing already converting is put at risk. ## The order to run them in Every change here is cheaper than a discount, and none of them touch your margin. But they are not equal, and if you run all six at once you will not know which one worked. **Run the framing and the default first.** Re-express the annual number so nobody is doing division at the worst possible moment, and look at what your toggle loads on and ask who decided that. Those two are the cheapest things on this list to ship, often a day of work each, and they move the plan mix. The plan mix decides your revenue per user, your payback period and your churn profile, so it decides everything downstream of the paywall. **Then the two placement changes.** Put the billing dates on the screen, and check whether your paywall is meeting people on Day 0 or waiting for a session that mostly never comes. Both are real work but neither is a pricing decision. **Trial length and the extra plan last,** not because they matter less but because they are the two you will want a clean read on. Look at your trial length and ask whether anyone currently at the company chose it, and against which category and which cash constraint. So if you are getting pushed to cut price, start at the top of that list. The discount is still there if none of it works. ## Common questions ### How can I increase paywall conversion without lowering my price? Change the framing and the defaults. Re-express the annual price as a monthly equivalent, default the plan toggle to annual, show the trial timeline including the billing date, place the paywall at the end of onboarding, choose trial length against your cash needs rather than by inheritance, and add a higher-priced plan so the target plan has a reference point. Each of these has published effect sizes behind it and none of them changes what you charge. ### Should my paywall default to monthly or annual? Default to annual unless you have a specific reason not to. Superwall's network data shows the plan mix moving from roughly 37% yearly to 63% yearly on that change alone, and 82% of users who do not close the paywall simply take whatever is pre-selected. The default is functioning as your recommendation whether or not anyone framed it that way. ### Is a long trial or a short trial better for a subscription app? They optimize for different things. Trials of 17 to 32 days convert at 42.5% against 25.5% for trials under 4 days, so length buys conversion rate. Short trials return ad spend faster, which is why RevenueCat has 46.5% of apps running 4 days or less. Choose based on whether your binding constraint is conversion or payback speed, and revisit the choice when that constraint changes. --- ## Subscription app benchmarks: the two funnel shapes your numbers reveal Source: https://ottergrowth.com/notes/the-benchmark-comparison Published: 2026-08-26 > Subscription funnels do not fail in a thousand ways. They fail in a small number of recognizable shapes, and the shape shows up fast once the numbers sit next to the right benchmarks. The clearest one is the AI-app signature: install-to-trial below benchmark, trial-to-paid at or above it, renewals fine, revenue per payer a little light. That reads as a monetization problem and it is a top-of-funnel problem. A blended cross-vertical median hides it, because that median is freemium-weighted and flatters some apps while punishing others. Give me eight of your numbers and I can usually tell you whether your problem is top of funnel, monetization, or retention. Not because I am clever. Because subscription funnels fail in a small number of recognizable shapes, and once the numbers sit next to the right benchmarks, the shape declares itself in about ten minutes. The hard part is not the diagnosis. It is having something honest to compare against. ## The eight numbers Before any shape appears, you need the same small set every time. Install to trial. Trial to paid. Install to paid. Early retention, whichever day marker your category uses. Renewal rate at the first billing period. Revenue per payer. The plan mix between monthly and annual. And refund or cancellation rate. That is the set. Eight numbers, and most teams can pull all of them in an afternoon. What makes them useful together is that each one isolates a different stage. Install to trial is about the paywall's first impression. Trial to paid is about whether the product delivered inside the window. Renewals are about whether it kept delivering. Revenue per payer is about the ladder you put in front of them. Read alone, each number is ambiguous. Read as a set, they form a pattern. ## The AI-app signature is the clearest one This is the shape I see most often right now, and it is the one that gets misdiagnosed most often too. Install to trial sits below benchmark. Trial to paid sits at or above it. Renewal retention is fine. Revenue per payer is a little light. Almost every team reads that as a monetization problem, because the revenue number is the one that hurts and it is the one the board asks about. It is a top-of-funnel problem wearing a monetization costume. The people who make it past the first screen convert well and stay, which tells you the product is doing its job. Not enough of them get that far. The fix is at the paywall's first impression. Presentation, trust, comprehension, what a stranger understands in four seconds. Not the price, and not the plan ladder. [Figure: Two subscription funnel signatures plotted against their benchmarks] Only the position relative to the line matters. Which metrics sit on which side is the diagnosis. ## The opposite shape, and the quarter it costs you The mirror image shows up just as often. Strong install to trial, weak trial to paid, weak renewals. That is a product-value problem. People are getting in easily and leaving anyway, which means the trial is not proving anything worth paying for. No paywall change is going to rescue it. I have watched teams spend a full quarter A/B testing a paywall that was converting fine, because the paywall is the surface everyone knows how to change and the product is the thing nobody wants to reopen. The tell that separates the two shapes is what happens after the money. If renewals hold, the product works and your problem is upstream. If renewals sag, the problem is the product, and every hour spent on the purchase screen is an hour spent making a leaky bucket prettier. ## Why a blended median hides both shapes Here is what makes this hard in practice. Both shapes look identical when you compare against a blended cross-vertical median. That median is freemium-weighted. It pools hard-paywalled apps with free-to-browse ones, weekly plans with annual ones, iOS with Android, AI products with utilities. Averaging across all of that produces a number that describes nothing in particular. It flatters some apps and punishes others, and you cannot tell from the number itself which one it is doing to you. Install to trial is where this bites hardest. An app with a hard paywall at the end of onboarding and an app that lets you browse for a week are not measuring the same event, so comparing their install-to-trial rates is arithmetic without meaning. ## Build the column, then read the shape What you need is not a better median. It is your own column, cut on the handful of dimensions that actually change the numbers. | Dimension | Why it moves the numbers | |---|---| | Access model | Hard paywall, soft paywall, and freemium produce completely different install-to-trial rates for identical products. | | AI or not | AI products carry a different comprehension cost at first impression and a different expectation about what the trial proves. | | Platform | iOS and Android differ on payer mix, price sensitivity and refund behaviour. A blended figure hides which one is dragging. | | Paywall placement | Onboarding versus in-app changes both the rate and which population is even measured. | | Vertical | Applied last, and it is the cut that most changes what "healthy" means for retention and renewal. | Get those five right and the comparison starts telling you something. Get them wrong and you are measuring your app against a population it does not belong to, which is how a team ends up fixing a number that was never broken. Then read the set as a shape rather than metric by metric. Metric by metric gives you a list of things that are slightly off, and every one of them looks like a project. The shape gives you one diagnosis and one place to start. ## The takeaway Pull the eight numbers. Build the column on access model, AI or not, platform, paywall placement, and vertical. Then look at which metrics sit above the line and which sit below, and stop reading them one at a time. If the people who get past your paywall convert and stay, your problem is that not enough of them get there. If they get in easily and leave anyway, no amount of paywall work will help you. Those are two different quarters of work, and the numbers tell you which one you are in before you commit to either. ## Common questions ### How do I know if my problem is top of funnel, monetization, or retention? Read the metrics as a pattern instead of individually. If install-to-trial is below benchmark while trial-to-paid and renewals are healthy, the problem is top of funnel, even when the revenue number is the one that hurts. If install-to-trial is strong but trial-to-paid and renewals are both weak, the problem is product value. Renewal rate is the tiebreaker: it holds when the product works and sags when it does not. ### Why are cross-vertical subscription benchmarks misleading? Because the blended median pools access models, platforms, price points and paywall placements that produce structurally different numbers for identical products. It is freemium-weighted, so it flatters apps with a low-friction entry point and punishes apps with a hard paywall, and the number itself gives you no way to tell which is happening to you. ### What metrics do I need for a benchmark comparison? Eight are usually enough: install to trial, trial to paid, install to paid, early retention at your category's day marker, first-period renewal rate, revenue per payer, monthly versus annual plan mix, and refund or cancellation rate. Each isolates a different stage, and together they form the pattern that identifies the failure shape. --- ## How to be the only human in the loop without being the bottleneck Source: https://ottergrowth.com/notes/the-one-word-decision-brief Published: 2026-08-05 > Being the only decision-maker doesn't make you the bottleneck. Being the only person who can do the analysis in front of the decision does. The fix is a fixed brief format: situation, recommendation, the specific weigh-in needed, ideally answerable in one word. Then protect the two things that genuinely need you, taste and the acceptance walk, and count how often you have to raise the same thing twice. Every founder I work with has some version of this problem. Everything routes through them, they know it's slowing the team down, and the advice they get is "delegate more." So they try, and one of two things happens. Either they delegate decisions they should have kept and spend a month undoing it, or they delegate nothing and stay the constraint. The framing is wrong. The bottleneck was never the number of decisions you make. A real decision takes seconds. The bottleneck is the work sitting in front of the decision: reading the thread, reconstructing the context, weighing the options, figuring out what's actually being asked. That's the expensive part, and it's the part that can move. I've been testing this at an extreme. I'm building a product with a fleet of AI sessions and I'm the only human on it, which means there's no one to absorb sloppy delegation on my behalf. If the interface between me and the work is bad, I feel it that day. ## Everything arrives in the same shape Nothing reaches me as a question. Everything reaches me as a brief with three parts. **Situation.** What is true right now, in a few lines, written for someone who has not been following. **Recommendation.** What the person or session bringing it to me thinks we should do, stated plainly, with the reasoning compressed to its load-bearing parts. **The specific weigh-in needed.** Not "thoughts?" but the exact call, scoped so tightly that the answer is a word. The third part is the one people skip, and it's where most of the time goes. "What do you think about the onboarding flow" is not a decision request, it's an invitation to do someone else's analysis. "We can ship the three-step version now or hold two weeks for the personalized one. Recommend shipping now and testing personalization after. Go or hold?" is a decision request. The best brief I got in the last month asked me to settle four things about a large rollout at once. I answered all four in four short phrases. Not because I'm decisive, but because someone had already done the thinking and left exactly four forks with a recommendation on each. [Figure: The anatomy of a one-word brief. Three stacked blocks feed one narrow decision point: situation, what is true right now; recommendation, what I think we should do and why; and the ask, one scoped fork. These lead to the founder's answer, a single word. The time cost sits in the first three blocks, not the answer.] Move the analysis, keep the call. The one-word answer is a test of the brief, not of the decider. That's the point worth holding onto. When an answer takes you twenty minutes, that's usually not a hard decision. That's a brief that made you do the work. ## Protect the two things that actually need you Compressing everything into briefs has an obvious failure mode: you start approving summaries of things you've never seen. So two categories stay unmediated. The first is taste. Whether a thing is good is not delegable and doesn't survive summarization. A report saying the copy reads well is not evidence that the copy reads well. The second is the acceptance walk. I go through the actual product myself, regularly, and that walk is the real gate. Not the tests, not the merge, not the report. In the same period where my decision input was a handful of short rulings, I also did six full walk rounds and sat with two teenagers using the thing. That's not a contradiction of the one-word interface, it's what makes it safe. The interface compresses decisions. It does not compress contact with the product or the user. ## Count how often you raise the same thing twice This is the habit I'd most want to hand to someone else, and it came out of being annoyed. Every correction I raise gets an ID and goes in a ledger, and nothing closes until it's verified in the product. Which means repeat-raises are visible. If I flag the same thing three times, that's counted, and it's a signal about the system rather than about the person. Here's the one that taught me. Something on screen felt wrong to me. It got "fixed" three times, and each fix addressed the animation of the element rather than the thing I was actually reacting to. I raised it a fourth time. That fourth pass finally measured it instead of guessing: an observer counted DOM mutations and found one element arriving as a single insertion with zero text mutations, while the line after it grew through 46. The subjective complaint had an objective signature the entire time, and three rounds of plausible fixes had never gone looking for it. Three rounds of "fixed" meant nobody had found the actual mechanism. The count is what surfaced that. If you don't track your own repeat-raises you will read that pattern as your team being sloppy, when it's usually your team solving a different problem than the one you described. The corollary rule: nothing a founder raises can be quietly parked. Killed on purpose is fine. Silently dropped is not, because you can't tell the difference from the outside and you stop raising things. ## This is the same thing I ask of a team None of this is AI-specific. The reason it shows up so sharply with a fleet of sessions is that the volume makes a bad interface unsurvivable within a day, where a human team can absorb a vague ask for a week before anyone admits they're stuck. The rule I hold my own team to is the same one: never bring a question without a recommendation. Not because the recommendation is always right. Roughly half get changed. But a recommendation forces the thinking to happen before the escalation, and it turns my job from analysis into judgment, which is the only part that needed me. ## The takeaway If you're the constraint on your team, don't start by giving away decisions. Start by fixing the shape of what reaches you. Situation, recommendation, one scoped fork. Keep taste and keep the walk. Track your repeat-raises and treat a third raise as a broken loop, not a lazy team. You want to be the person who answers in one word, because everything upstream of that word already got done properly. ## Common questions ### How do you delegate without losing control of quality? Delegate the analysis, not the acceptance. Everything comes to you as a brief with a recommendation and a scoped decision, so you're spending your attention on judgment instead of reconstruction. Then keep two things unmediated: subjective quality calls, which don't survive being summarized, and a regular hands-on walk through the real product, which stays the actual acceptance gate. ### What does a good decision request look like? Three parts. The situation in a few lines written for someone who hasn't been following, a clear recommendation with the reasoning behind it, and one specific fork scoped tightly enough to answer in a word. If the ask is "thoughts?" it isn't a decision request, it's a request that you do the analysis. ### Why track how often you repeat feedback? Because a repeat is diagnostic. If the same issue comes back a third time, the usual cause is that each fix addressed a plausible mechanism rather than the real one, and nobody measured. Counting the raises makes that visible early, and it separates a system problem from a people problem, which are handled very differently. --- ## Running parallel AI coding agents: how I got to 400 commits a week solo Source: https://ottergrowth.com/notes/building-with-a-fleet-of-ai-sessions Published: 2026-07-30 > In seven weeks I pushed 1,439 commits and shipped a working AI product, alone, nights and weekends, alongside my consulting work. Output went from about 70 commits a week to about 400. That 6x did not come from better prompts. It came in three stages: I stopped being the only reviewer, then I stopped improvising and turned the whole thing into a system with governing plans, settled decisions, and a review loop that runs itself. Peer review between AI sessions got me roughly 2x. Systematising it got me the rest. Almost every guide to building with AI coding agents is really a guide to prompting. Prompting stopped being my constraint weeks ago. The constraint is trust. You can only ship as fast as you can believe what comes back, and for a while my only mechanism for believing it was reading all of it myself. That works, right up until it doesn't. Here's what actually changed, measured from git rather than remembered. [Figure: Three stages of build velocity over seven weeks. Stage one, weeks one to three, about 70 commits a week, when the founder reviewed all the work himself. Stage two, weeks four and five, about 150 a week, once Claude sessions reviewed other sessions' work. Stage three, weeks six and seven, about 400 a week, once a strategic plan, ratified decisions and a review loop on a timer were in place. The two-week trailing average runs 72, 62, 119, 152, 310, 396.] Two-week trailing average of commits per week, weeks two through seven. Derived from git, bucketed by ISO calendar week. ## Stage 1: I was the bottleneck, and quality was never the problem For the first three weeks I read every line of AI output myself before anything merged. Every change, by hand. The work was good. That's the part people get wrong about this stage (and it's why "just review everything" sounds like such reasonable advice). Nothing was broken, nothing was sloppy, and the review caught real things. The problem was arithmetic: I could only move as fast as I could personally read, which turned out to be about 70 commits a week. Weeks one to three ran 73, 71 and 52. Week three was the slowest stretch of the entire project. That is not a coincidence and it is not a dip in effort. It is what a human review bottleneck looks like when the thing behind it is capable of producing far more than the human can absorb. ## Stage 2: sessions reviewing other sessions, which roughly doubled it The first unlock was giving up the idea that I had to be the reviewer. I started having AI sessions review other sessions' work, with my input on what to look for, and surfacing only the calls that genuinely needed me. Output roughly tripled week over week and settled at about 150 a week. This is where a lot of people stop, and I understand why. It feels like the whole answer. Sessions can critique each other, the obvious errors get caught, and your throughput goes up immediately. But it is only half a system, and the half it is missing is the one that compounds. ## Stage 3: the part that actually produced the 6x The second unlock was stopping the improvisation. Up to that point every session got a hand-written brief, and I was the only thing holding the shape of the work in my head. Turning that into a system took six pieces. If you're starting Monday, build them in this order. The written record of decisions first, because everything downstream reads from it. Then one session in charge. Then the mechanical checks. The rest can wait. [Figure: A six-step loop for building with AI sessions. One, you use the product and say what is wrong in plain sentences. Two, your calls become written rules every session has to follow, 62 so far. Three, one session in charge reviews every change and wakes itself every half hour. Four, two to four AI sessions build at once, seven at peak, each isolated. Five, nothing is believed until a command proves it. Six, every promise gets a row and an owner, or the whole pass fails. It does not replace your judgment, it stops wasting it.] The six pieces and how they hand off to each other. The governing session is the hub because that is where every change passes through. **Two documents everything obeys.** A strategic plan carrying the product thesis, the market read, and what actually makes this different. Under it, an execution roadmap. Both are versioned in the repo and both get revised constantly, roughly 80 times each so far. **A written record of settled decisions.** Sixty-two ratified rulings, each one a call I made that no session is allowed to relitigate. The format matters: when I change my mind, the entry gets rewritten in place rather than patched with an exception. I learned that the hard way. A register full of amendments-to-amendments is unreadable, and unreadable law gets ignored. **One session in charge.** It writes the brief for each work stream, reviews every change before it lands, merges, and spawns follow-on work when a session turns something up. It runs on the strongest model available because this is the seat where judgment actually happens. **A review loop that runs itself.** That session re-arms its own watch on a 30 to 45 minute cycle. It re-derives state, checks the running streams, and most passes send nothing at all. The point is not constant intervention, it is that nothing sits unnoticed for a day because I got busy. **Model tiering per stream.** The top tier for thinking, evaluation and the review seat. A mid tier for execution. A cheaper tier for well-specified mechanical work. Every brief carries an explicit model recommendation with a one-line reason, so the choice is deliberate rather than habitual. **Prove one, then fan out.** Nothing scales until a single instance is proven end to end, by hand, by me. The last large wave went one pilot, then three sequential provers each de-risking a different pattern, then parallel batches, then eighteen of eighteen done. Fan out before the pilot is proven and you find the flaw fourteen copies later. ## The number that justifies all of it About half of what comes back from a working session has something wrong or stale in it. Not obviously broken. Confident, well-organised, and off in some detail you have to go looking for. Run multiple AI coding agents in parallel and you do not get parallel work, you get parallel confidence. That single figure is the entire argument for a review seat, and it is why the checks are mechanical rather than cultural. Thirty-two named failure modes live in the repo, each one ending in a command to run rather than advice to remember. Because here is the thing I keep relearning: knowing a failure mode by name does not stop you committing it. Sessions have written up a failure mode and made its sibling mistake two hours later. Guards, not goodwill. The clearest example: a change that would have silently deleted every second tip in a core flow. No type error. No failing test. It was caught only by deliberately reinserting the old defect and watching whether the guard fired. If the guard had been a code review comment instead of a command, that ships. ## What it does not do It does not replace my judgment. It stops wasting it. The system's real job is surfacing the handful of calls that genuinely need me, in a fixed shape: situation, recommendation, and the specific weigh-in needed, ideally answerable in one word. Everything else gets decided against the written record without me. Two to four streams run at once, seven at peak. I am still the only human on it. I make more decisions per week than I did in stage one, and I read almost none of the code. ## The product this built It is called Build Me, it is free, and it is at [build-me.org](https://www.build-me.org/). It helps people work out which career actually fits them by doing the work rather than reading about it, through day-in-the-life simulations and real projects run with an AI assistant. If you try it, tell me where it is wrong. And if you know someone working out their direction, a teenager, a new grad, anyone mid-switch, I would genuinely like to interview them. ## Common questions ### How do you stop parallel AI coding agents from breaking each other's work? Isolated git worktrees, one change per session, and a single reviewing session that holds the merge. The isolation stops them writing over each other mechanically; the review seat stops two individually-correct changes from combining into something broken. ### Does this only work because it is a side project with no users? Partly, and that is worth saying plainly. Pre-launch means no production incidents competing for the review seat's attention. But the constraint the system solves, more capable output than one person can verify, gets worse with scale, not better. ### What is the honest gap in the measurement? Wake cycles are not logged as first-class events. They are reconstructable from commit clusters and close-out notes, but there is no telemetry, and the same is true of model routing. So "it wakes every half hour" is configuration plus scattered receipts, not an instrumented log. The commit and pull request counts are exact. ### Would you run this on a team instead of alone? The layer that would change is the decision register, which currently only has to satisfy me. With more humans it needs a ratification path that is not one person's judgment. Everything else, the review seat, the mechanical gates, the ledgers with owners, gets more valuable with more parallel work, not less. --- ## The new bottleneck in product is judgment, not research Source: https://ottergrowth.com/notes/the-new-bottleneck-in-product-is-judgment Published: 2026-07-23 > The old center of gravity in product, months of deep research and bespoke design to guess at a solution, is losing its relative importance. Two forces did it: AI made building cheap and fast, and most product problems now have proven comparables and known dynamics. The bottleneck moved from gathering research to judgment. That is more opinionated understanding of your user, not less, and it does not take months to get there. The most valuable skill in product used to be research. It isn't anymore. And I want to be careful with that sentence, because the obvious misreading is "understanding users matters less." It doesn't. The bottleneck moved. Ten years ago the hard part of product was guessing. You ran months of deep user research and bespoke design to arrive at a solution nobody had built before, because there was nothing to copy and no cheap way to find out if you were right. Building was slow and expensive, so being wrong was slow and expensive. Research was the insurance policy against a build you couldn't afford to waste. Two things changed that. ## Two forces moved the center of gravity The first is that AI made building cheap and fast. A working prototype that used to be a two-week sprint is now an afternoon. When the cost of trying a solution collapses, the value of a long research phase whose whole job was to avoid a wasted build collapses with it. You can just build the thing and look. The second is quieter and matters more. Most product problems now have proven comparables and known dynamics. Onboarding, paywalls, referral loops, marketplace liquidity, activation, retention curves, these are not blank pages anymore. There is a decade of public teardowns, benchmark data, and pattern libraries for almost any consumer surface you are working on. Ten years ago you had to guess how to solve a product problem. Usually you don't. Someone has solved a version of it, and the shape of the answer is knowable before you write a line of code. So if building is cheap and the solutions are mostly knowable, what's left to be hard? ## The bottleneck moved to judgment It also changes who is scarce. When building is the constraint, engineers are the constraint. When building is cheap, engineering gets commoditized, design does to some extent too, and the product manager gets more important than they have been, because the thing nobody can buy off the shelf is the quality of the calls. The scarce skill is no longer gathering research or guessing a solution. It's deciding. What the problem actually is. Who exactly you are building for. What the solution is. What people would use instead, including nothing. And which two things you are betting on being better at. That is not less understanding of your user. It's more, and a more opinionated kind. Cheap builds and known patterns don't make the thinking optional. They raise the price of thinking badly, because the constraint on your product is now the quality of your calls, and not the speed of your hands. You can hold a sharp, well-argued point of view about your user and your market in a week. The old process treated a long calendar as proof of rigor, and it isn't. Here is the formula I actually run now, before building anything new. Five questions, and every one of them is a judgment call rather than a research task. [Figure: The ICP formula, five questions in order: what is the problem, who is the ICP, what is the solution, what are their alternatives including doing nothing, and what are the two differentiators.] The ICP formula. All five are judgment calls, and you can have honest answers to them inside a week. **1. What is the problem?** Not the feature request, and not the thing the loudest customer asked for. The problem underneath it, stated plainly enough that somebody could disagree with you about it. Most product arguments are two people solving different problems and not noticing. **2. Who am I building it for? Who is the ICP?** Not a persona slide. Specific enough that you could name ten of them by tomorrow, and defensible enough that you would bet the roadmap on it. If you cannot name ten, what you have is a category rather than a customer. **3. What is the solution?** Naming the thing honestly matters here, and it is the one people get wrong. A product is rarely one clean category. It is usually a combination of a few, and naming that combination is a real decision rather than a positioning exercise. Get it wrong and everything downstream aims at the wrong target. **4. What are their alternatives, including the competitors and including doing nothing?** Doing nothing is the one people leave off, and it is usually the market leader. With comparables sitting right there you are not guessing at what someone could use instead. You can go and look. **5. What are the two differentiators?** Two. Not seven. The ones that make you meaningfully better at the one job in question. With comparables visible to everyone, differentiation is a choice rather than a discovery, and the choice is load-bearing, because it changes both the product you build and the user you build it for. If everything is a differentiator, nothing is being bet on. Answer those five honestly and the build is the cheap part, which is the whole argument of this piece. This is the same frame that turns up in [what product-market fit actually means](/notes/what-product-market-fit-actually-means) as the four blanks. Not a coincidence: measuring PMF and deciding what to build next are the same judgment asked at two different moments. ## How to actually prioritize between problems Question 1 is where teams stall, because every problem looks important when you list them out. Size them instead. Two axes. **Relevancy:** what share of your users actually have this problem. **Severity:** how much it blocks them or matters to them. The trick on severity is to force the ranking, a 1-to-5 scale or a forced-rank or MaxDiff, so users can't rate everything "very important." If people are allowed to call everything a priority they will, and the ranking you get back tells you nothing. [Figure: A two-by-two matrix of user problems by relevancy and severity. High relevancy and high severity is build here first. High severity, low relevancy is niche fixes. High relevancy, low severity is nice to have. Low on both is ignore.] Relevancy times severity. The top-right quadrant is where the metric actually moves. Then get it direct. Poll your users with the bluntest possible question: why do you come back, or why don't you. You will learn more from a hundred of those answers than from a month of inferred journey mapping. Run probing tests and small bets across a few of the top areas, watch what actually moves the needle, keep collecting feedback, and hone in on the handful of problems that matter. The prioritization isn't a one-time ranking, it's a loop you tighten. ## This isn't theory. It's how the work goes now. A returning client came back recently wanting more user control in the product, more options, more configurability. Comparables and a few JTBD interviews said the opposite. Their users didn't want more control, they wanted it done for them. So we cut the option menu and fronted a single core path, one AI-driven flow that did the work instead of asking the user to. Activation doubled. The insight was the entire job. The build was an afternoon. I'm running the same playbook on my own product right now, a zero-to-one build I'm testing with real users. I named my hypotheses about the space up front, mapped them to product combinations and to comparables that already exist, and I'm testing them with the actual target users to prove or disprove each one fast. There was no six-month research phase, just a point of view and then evidence against it. And it holds outside consumer apps. Diagnose a marketplace that's short on supply, and you don't start from a blank page. Marketplace liquidity has known patterns. You can reason about it as a product-identity problem, a trust-first marketplace, say, and ask where its gap is versus the incumbent everyone compares it to, rather than commissioning net-new research to rediscover dynamics the category already understands. ## The takeaway Engineering used to be the constraint. You could not build fast enough, so you had to be very sure what to build before you spent months on it, and certainty was expensive because building was expensive. Building is cheap now and most solutions are knowable, so the edge is not in the research or the build anymore. It is in the calls, and the calls are five: what the problem is, who the ICP is, what the solution is, what the alternatives are including doing nothing, and which two things you are betting on being better at. That is more opinionated understanding of your user, not less, and you can get there in a week if you are rigorous instead of slow. The research was never the moat. The opinion is. ## Common questions ### Does user research matter less now? No, the opposite. What changed is the bottleneck, not the value of understanding users. You no longer need months of research to guess a solution, because building is cheap and most problems have proven comparables. So the scarce skill moved to judgment, which requires a sharper and more opinionated understanding of your user than a generic research phase ever produced. ### How do you decide which user problems to work on? Size each problem on two axes: relevancy (what share of users have it) and severity (how much it blocks or matters, asked with a forced-rank or MaxDiff so users can't call everything important). Prioritize the high-relevancy, high-severity quadrant, confirm it by asking users directly why they do or don't come back, then run small bets and keep the ones that move the metric. ### What replaces the long research phase in product work? A fast, evidence-backed point of view, answering five questions: what the problem is, who the ICP is, what the solution is, what the alternatives are including doing nothing, and what your two differentiators are. Map those against known comparables, then build a cheap version and test it with your ICP to prove or kill each one quickly. It is a faster process, not a shallower one. --- ## The best B2C teams don't plan a launch Source: https://ottergrowth.com/notes/dont-plan-a-launch Published: 2026-05-26 > There is no launch moment any more. The advice founders still get, build a waitlist, grow the email list, pick an announcement day, line up press, is antiquated and it does not drive much. The best B2C teams ship early, put a price on it, and iterate until people actually love it. That is when you get a launch worth having. The best teams I've worked with don't plan launches. They're animals about iteration. Product Hunt. A PR push. Investor buzz. Those are headlines, not a growth strategy, and investors love to fantasize about a splashy launch day. Here's how it actually goes. You spend three months planning launch day. You get your spike. A week later the graph is exactly where it was. The teams that win build a system that compounds from day one, and if they bother with a launch at all, it's a bonus sitting on top of that. ## The launch advice is the problem, not the execution Build a waitlist. Grow the email list. Pick the announcement day. Polish until it's ready. Line up press and influencers. That's the advice founders still get, and it's antiquated. It isn't that teams execute it badly. It's that every piece of it is doing less than the person recommending it thinks. An email list does not convert strangers. Someone handed over an address months ago because a landing page asked nicely. That is not a buyer, and emailing them on the day you go live does not make them one. You are counting an audience you don't have. Press and influencers won't save it either, because nobody covers a v1. There's nothing to say about it yet. Which is the real problem underneath all of it: on the day you go live, the product isn't good enough for anyone to talk about. That's just what a first version is. So you spend three months engineering one good day, on top of a product that hasn't yet earned the conversation you're trying to manufacture. ## A launch spike does not compound The launch-day model treats growth as an event you can schedule. Consumer subscription growth is not an event. It is a compounding loop, acquisition into activation into retention into revenue, and it gets a little better every week you run it. A launch spike borrows a burst of attention and gives almost nothing back once it fades. The loop keeps producing, and what it produces this week sits on top of what it produced last week. That is the real difference between the two graphs below. One is a moment. The other is a machine. [Figure: A launch produces a spike that decays back to baseline. A compounding system starts small and grows steadily past it over time.] The launch is the grey spike: a burst that decays. The system is the ember curve: small at first, then past the spike and still climbing. ## Ship early, and launch with a paywall Ship before you're comfortable. Real users in your real flow will teach you more in a week than interviews, friends and internal reviews teach you in a quarter. You want to see the metrics. You want to see where people drop. You want to see who comes back, and why. And launch with a paywall. Not because you will make real money on day one, but because it is the fastest way to get signal on what people will actually pay for and how far you are from profitable acquisition. That number shapes everything else. If budget is tight and you genuinely need volume first, a one-to-three month free window can make sense. ## Then point your acquisition energy here Once the product is live and the paywall is teaching you something, focus acquisition in this order: a strong app store listing, a clean signup and onboarding flow, a dialed-in paywall and activation experience, and then a paid acquisition test. Meta and Google are your most predictable starting points. Influencer marketing, organic social, SEO, and Reddit are all worth building, but know what they are: six-month to one-year plays. They won't give you the volume you need to learn right now, and they're often more expensive and more operationally heavy than just running paid. The organic engine comes from product delight. Iterate toward that. The rest follows. ## The weekly gains are the ones that add up None of this produces a screenshot-worthy spike, and that is the point. A one percent improvement in activation this week, a better paywall test next week, a tighter creative loop the week after, none of it trends on launch day. But it stacks. A year of stacking is what an actual business is made out of, and the launch day is a good story you tell later. ## The takeaway Stop planning a launch and start building the system. Ship early, put a price on it from day one, and improve the loop every single week. The spike will feel like progress and then be gone. The slow weekly work is the thing still paying you a year from now. And you do get a launch out of it. Just not the one you were told to plan. Announce it once people already love the product, when there's finally something worth covering and an audience that will actually pass it on. That one is a real launch, and it is the only one worth putting three months into. ## Common questions ### Should a B2C app skip launching entirely? Not exactly, but launch later than you were told to. A waitlist, an email list and an announcement day put the moment before the product is good enough for anyone to talk about. Ship early, iterate until people love it, and announce it then. That launch is worth having; the one you plan for a v1 is not. ### What should I focus on instead of a launch? The compounding loop: get the product to real users early, start testing the paywall from day one, and improve activation, retention, and monetization every week. ### Why is the paywall part of a launch strategy? Because it moves lifetime value across your entire base, so every week you test it, the whole system gets more valuable. Turning it on early is where the compounding starts. --- ## Referral landing pages: referred users are worth 3x, and this flow keeps them Source: https://ottergrowth.com/notes/referred-users-worth-3x Published: 2026-04-13 > The share moment is the window between one person tapping Share and another person accepting, and it is where most products squander their most valuable users. A referred user is worth about 3x a cold one (Branch.io) and then lands on a homepage. Convert the moment instead: lead with the connection, show the actual thing that was shared, social proof, one low-commitment CTA, all on one screen. The most valuable user you will ever get is a stranger one of your customers just vouched for. Most products convert that person with a page that ignores the vouching entirely. ## The share moment is the whole thing The window I care about here is narrow. One person taps Share. Another person accepts. Everything that decides whether you get that second user happens inside that window, and it is the window teams put the least work into. Referred users generate about three times the revenue of cold ones. That's Branch.io's number, and it is a good reason to build referrals. It is a better reason to obsess over the screen most teams treat as an afterthought: what the person on the receiving end sees after tapping a friend's link. Most of those are built like a brochure for a stranger, and then everyone acts surprised when the warm user bounces. Think of it like an ad. You get one quick chance to hook them, or they're gone. Here's the flow I use with clients, and the psychology behind why it works. [Figure: A referred user is worth about three times the revenue of a cold user, according to Branch.io.] Source: Branch.io. ## Send them to web, with a content preview Before the landing page even loads, one routing decision matters. Send referred traffic to a web page with a preview of the content, not straight to the app store. Branch.io puts the difference at three to six times the click-to-install rate. The app store is a cold, generic destination. A web preview keeps the warmth the referrer created. ## Lead with the connection, not your brand The first thing a referred user should see is the person who referred them. "[Name] wanted you to see this," framed like it came directly from them. This is the most important element on the page, because the only reason this user showed up is the person who sent them. Peer trust is the conversion engine. Your logo goes small and out of the way. It's not why they're here. [Figure: The referral recipient landing page template: a small logo, the referrer connection line as the hero, a content preview, a headline specific to the shared content, a one-line value statement, a low-commitment Try Product CTA, and a social proof line. Everything on one screen.] The referral recipient landing page template I use with clients. Every element earns its place, and the whole thing fits on one screen. ## Show the actual content Show the specific thing the referrer experienced, not a generic pitch. A page built around the real shared clip beats a generic landing page by a wide margin. Give them a taste of the value before you ask for anything, a moment of delight that mirrors what activation actually feels like. Keep the headline specific to that content so you don't break the personal connection the referrer built, and keep the value statement to one line: what they get from the product. ## Social proof, one CTA, one screen Close with a rating and a single low-commitment call to action. Try Product, pointing at the app store. That is it. Try Product beats higher-commitment CTAs every time. Add the social proof line, a 4.9 and the number of people who love it, because community validation seals what the referrer started. The whole page fits on one screen. No scrolling. No feature list. No pricing, because they haven't seen the thing yet, so don't ask them to price it. Referral pages fail when they try to do too much. Every time. They work when they stay on the one thing already running in your favor, which is the trust someone else built for you before the click. Remind them why they're here, give them one good moment, and get them to tap through. The app store does the rest. ## The takeaway A referred user is worth roughly three times a cold one, but the program is not the point. The share moment is: the window between someone tapping Share and someone else accepting. Lead with the connection, show the exact thing that was shared, one CTA, one screen. Do that and you keep the value the referrer already earned for you. Clutter it and you hand it back. You pay real money for strangers. This one showed up already trusting you, and most teams hand them a homepage. ## Common questions ### Why are referred users worth more? Trust transfers from the person who referred them, so they arrive warmer and convert better. Branch.io puts a referred user at about three times the revenue of a cold one. ### Should referral traffic go to the app store or to a web page? Send it to a web page with a preview of the shared content. Branch.io measures that at three to six times the click-to-install rate versus dropping people straight into the app store. ### What is the most important element on a referral landing page? The referrer connection. Naming the person who sent them, front and center, is what converts, because peer trust is the reason they showed up. --- ## Six paywall A/B tests to run first (your paywall is the highest-leverage experiment you have) Source: https://ottergrowth.com/notes/paywall-highest-leverage-experiment Published: 2026-04-07 > Your paywall moves lifetime value across your whole base, so it deserves a disproportionate share of your testing budget. Start by making annual the only visible plan, show the trial timeline, and lead the CTA with free. Six tactics below, in the order I would test them. Most teams spend their experimentation budget in the wrong place. They test onboarding tweaks and button colors and leave the paywall alone, because it feels risky to touch the thing that takes the money. So they polish the button color and leave the pricing untouched. That is not caution. That is flinching. The paywall is the highest-leverage surface you have. The difference between sixty percent of users buying annual and thirty percent buying annual can be the difference between a profitable business and one that never gets off the ground. One number, moved on one screen, changes the math for every cohort that follows. [Figure: Same product and same price. When 30 percent of users choose annual the business struggles; when 60 percent choose annual it becomes profitable. The plan mix decides the outcome.] Same product, same price. The share of users who pick annual decides whether the business works. That leverage is why the paywall deserves structured testing, not the occasional guess. Here are the six tactics I have seen actually move the needle, in the order I would test them. ## 1. Make annual the only visible plan Do not just highlight annual. Make it the only plan visible, with monthly and weekly tucked behind a View All Plans submenu. Most users take the path of least resistance, so make that path the one with the highest lifetime value for you. This is the first thing I test on any paywall, because the plan people see is the plan they buy. And however you price it, show the annual plan broken down to its monthly or weekly equivalent, so the comparison is obvious. ## 2. Show the trial timeline on the paywall Tell users exactly what will happen and when. Day one, the trial starts. Day five, we remind you. Day seven, you are charged. Then actually send that reminder. It reads as a courtesy, it removes the quiet fear that a free trial is a trap, and it converts better than the vague version every time. [Figure: A trial timeline shown on the paywall: day one the trial starts, day five a reminder is sent, day seven the user is charged.] Put this on the paywall itself. Certainty converts better than a vague "free trial" promise. ## 3. Test weekly plans RevenueCat has weekly plans at 55% of all app revenue, up from 43% two years ago. Annual captures your high-commitment users; weekly converts the low-commitment ones who would otherwise never start. You want both doors open. The plans you are nervous to show are often the ones reaching demand you are currently leaving on the table. ## 4. Nail the CTA and the reassurance around it Lead with free. Start free trial beats Subscribe every time, because it names what the user is actually agreeing to right now. Put star ratings or a short testimonial directly on the paywall, and add one line at the bottom: cancel anytime in your settings. Small thing. Real impact on anxiety. ## 5. Bring excitement back up before you ask Personalization questions are necessary and they flatten people out. By the time someone has answered eight questions about their goals and their experience level, the feeling that made them download the thing has drained out of them, and that is the state they arrive at your paywall in. So put a high moment in front of it. A congratulations, a result, something that shows them what they just built. Then social proof and trust, before the purchase decision lands. You're not hiding the paywall. You're making sure they reach it feeling something other than processed. ## 6. Test enabling and disabling the trial Let the user choose whether to turn the free trial on. It hands them a sense of control at the exact moment they feel cornered, and it works as a small commitment. For the ones who opt out of the trial, you get an easy lifetime-value win. ## One more, and it outranks all six Where the paywall sits decides how much any of the above matters. The minimum bar is that the user understands what the product is and what they're going to get from it. If they don't, they will never start a trial, and no amount of framing rescues it. Every tactic on the list above is a way of improving a decision the user is ready to make. None of them create readiness. I watched a client move their paywall much earlier. Trial starts jumped, which is exactly what you would predict and exactly what the dashboard celebrated for about a week. Trial conversion fell, because we'd pushed people into a trial before they were ready and they never saw the value. Net worse. That is the trap in this particular metric. Moving the paywall earlier always looks like it worked, because the number it moves first is the number that responds fastest. You have to wait for the conversion cohort to catch up before you know anything. Timing is the lever underneath all the others. ## The takeaway Stop treating the paywall as the fragile thing you do not touch. It is the one surface where a single test moves lifetime value across your entire base, so it should get more of your budget, not less. Start with annual as the only visible plan, work down the list, and run it like a program instead of a screen you set once and forget. And before any of it, check that the user reaching the paywall actually understands what they're being asked to buy. ## Common questions ### What is the highest-impact change to a subscription paywall? Make annual the only visible plan, with monthly and weekly behind a submenu. Most users take whichever plan is in front of them, and annual carries the most lifetime value. ### Should we offer a weekly plan? Test it. Weekly now makes up more than half of app revenue, around 55 percent per RevenueCat, because it converts low-commitment users that annual and monthly never reach. ### Where should the paywall go in the flow? Late enough that the user understands what the product is and what they'll get from it. That is the minimum bar, and it outranks every framing tactic. Moving the paywall earlier reliably lifts trial starts and can quietly lower trial conversion, so wait for the conversion cohort before calling it a win. ### Does showing the trial timeline hurt conversion? No, it lifts it. Telling users exactly when they will be charged, and reminding them before it happens, removes the fear that the trial is a trap. --- ## Meta ads targeting is dead. Creative is the targeting now Source: https://ottergrowth.com/notes/targeting-is-dead Published: 2026-03-31 > Targeting is dead, and creative is the targeting now. You used to use targeting settings to tell Meta which audience to go after; it now matches on the creative itself, which is why it pushes everyone toward Advantage+ audiences and demotes what you typed to a suggestion. So you reach a different audience by making creative that speaks to that persona, and by spending enough to give the algorithm something to optimize on. Restrictions just make it less efficient or point it the wrong way. Meta now treats the targeting you type in as a suggestion. That is not a quirk of the interface. That is the platform telling you it can find these people better than you can, and it is right. If I open your Meta account and the first thing I see is a stack of interest audiences and lookalike sets, I already know what happened. Someone sold you on the old way. That is where paid accounts used to be won. It is not where they are won anymore. The lever moved, and most teams are still pulling the old one. ## The algorithm targets better than you do The algorithm is good enough to find your people without you telling it who they are. Broad targeting plus a strong conversion signal now beats hand-picked audiences almost every time, because the platform watches thousands of signals on every user and finds the likely converters far better than your guess about which interests matter. Meta knows this, and it has stopped being subtle about it. That is what Advantage+ audiences are: you can still type an audience in, and the platform treats what you typed as a suggestion rather than a constraint. Read that as the tell it is. The company with the data is telling you its algorithm can find these people for you. When you box the algorithm into a narrow interest audience, you are handing it a worse version of the job it already does automatically, and starving it of the scale it needs to learn. Give it too many restrictions and you do one of two things: make it less efficient, or put it on the wrong path entirely. So the manual audience-building most teams still obsess over is not just low-value. It actively gets in the way. [Figure: Then: narrow hand-picked audiences and a few ads. Now: one broad audience with lots of creative, and the algorithm finds the people.] The work moved from choosing who sees the ad to making enough creative for the algorithm to work with. ## The creative is the targeting Here is the shift that matters, and it is an inversion rather than a downgrade. You used to use targeting settings to tell the algorithm which audience to go after. Now the creative does the finding. Which means the lever is still there. It just moved. If you want to reach a different audience, you make creative that speaks to that persona, and the ad goes and finds them. A specific ad speaks to a specific person, and the people it resonates with are exactly the audience the algorithm learns to chase. The hook, the visual, the first three seconds: that is your audience selection now. So the question stops being which audience do I pick, and becomes which personas am I actually speaking to, and how many distinct angles can I get in front of the machine. That reframes the whole job. You are not an account manager tuning audiences. You are running a creative operation. The other half of the reframe is spend, and it gets skipped. The algorithm needs enough data to optimize on. An account that never gets out of the learning phase is not going to find anyone for you, however good the creative is, and no amount of creative volume compensates for a budget too thin to produce a signal. ## What actually moves the needle: three things Once you go broad, the account is won on three things. Most teams only do the first one. ### 1. Creative volume and velocity Not one polished ad, and not fifteen variants of the same angle. Dozens of hypotheses running at once, across different personas, angles and hooks. Different people, different problems, different hooks. More variants means faster learning means better results, and the team shipping thirty concepts a week beats the team shipping three and arguing about audiences. ### 2. Clean data signals This is the one most teams get wrong from the start. Your front-end and back-end events have to fire correctly and pass as many user properties back to the platform as possible. Good creative with broken data is a leaky bucket: the algorithm can't optimize toward outcomes it never sees. ### 3. End-to-end automation This is where the best programs pull away from everyone else. Performance data feeds automatically into analysis. Winners and losers get identified without manual reporting. Results feed into new creative briefs, new variants get generated and launched, and the loop repeats as fast as possible, with AI at the center of it. One person running this system beats a full agency running interest audiences, every time. [Figure: The creative loop: produce many concepts with AI, launch broad, read the signal, cut losers and scale winners, then produce again. Volume and velocity.] The account is won by turning this loop faster than the competition, not by finding a cleverer audience. ## What to actually do Go broad. Give the algorithm room, a clean conversion signal to optimize against, and enough budget to get out of the learning phase. Then put your real energy where it now belongs: a creative pipeline that produces a high volume of distinct angles aimed at distinct personas, a testing rhythm that cuts losers and scales winners quickly, and AI in the loop to make both faster. Stop grading yourself on how clever your audiences are. Start grading yourself on how much good creative you ship and how fast you learn from it. ## The takeaway Creative is the targeting now. That is the whole thing. You used to tell the platform who to go and find. Now you make something that speaks to a particular person and the platform goes and finds them, which is why it demotes what you typed to a suggestion and pushes you toward Advantage+. Give it a broad target, a clean signal and enough spend to learn on, and move your effort to creative aimed persona by persona. Restrictions do not make it more accurate. They make it less efficient, or they put it on the wrong path. ## Common questions ### Should I still build custom audiences and lookalikes on Meta? Mostly no. Broad targeting with a clean conversion signal now beats hand-picked audiences, because the algorithm optimizes on far more signal than you can. Put that effort into creative instead. ### If targeting is dead, what actually decides paid performance? The creative, because the creative is now doing the targeting: you reach a different audience by making something that speaks to that persona. Behind that, three things. Creative volume and velocity across distinct personas, clean data signals so the algorithm can optimize toward real outcomes, and end-to-end automation that turns results into the next round of creative. You also have to spend enough for the algorithm to have data to optimize on. Great creative on broken data, or on a budget too thin to learn from, is a leaky bucket. ### How much creative do I need to ship? As much as you can produce and honestly read. Volume gives the algorithm angles to work with; velocity lets you cut losers and scale winners fast. An AI-assisted pipeline is how you do both faster than competitors. --- ## Subscription app pricing and free trial structure, the simple version Source: https://ottergrowth.com/notes/subscription-pricing-and-trial-structure Published: 2026-03-19 > People convert on a trial only after they experience the value, and how long that takes is set by the product: day one for entertainment, so 3 to 7 days works; a week or two for wellness and education; around 30 days for health and fitness, where physical results have to show. Balance that against intent and urgency, which is also why lifetime memberships exist. If you are early, optimize for learning rather than price. Entertainment can get away with a 3-day trial. Health and fitness usually can't do it in under 30. Same mechanic, completely different number, and almost nobody picks theirs on purpose. Most people overcomplicate pricing. They run willingness-to-pay surveys and design five-tier plans before they have a hundred paying users. There are proven patterns. Start there. ## 1. Trial or no trial Offer one. Full stop. Trust is too low without it. The question isn't whether to trial, it's how long. And there is only one question that sets the length: how long does it actually take someone to experience the value? Nobody converts on a trial until they have felt the thing working. That is the whole mechanism, and it is why trial lengths play out the way they do by category rather than by fashion. **Entertainment.** You can argue the value lands on day one. Somebody watches something and they either liked it or they didn't. This is why 3-day and 7-day trials work here. **Wellness and education.** A week minimum, and 14 days is often closer to right. Learning something or building a practice takes more than one sitting to feel like anything at all. **Health and fitness.** You need actual physical results to show up. 30-day trials are a lot more common here, and that is not generosity. It is just how long it takes to see anything. Get it wrong in either direction and you either charge before they've felt anything, or you hand the value over and lose them right when they were ready to pay. You do have to balance it against intent and urgency, though, because the two pull opposite ways. A longer trial gives someone more room to feel the value and more room to forget why they signed up. RevenueCat has trials of 17 to 32 days converting at 42.5% against 25.5% for trials under 4 days, and also has 46.5% of apps running 4 days or less anyway, because short trials pay ad spend back faster. Both of those are real answers to different constraints. Picking one by accident is not. That same balance is why lifetime memberships exist. They work when the thing is something you want to invest in for the rest of your life: personal investment, wellness, fitness. Nobody buys a lifetime membership to something they expect to be finished with. [Figure: Trial length set by how long the value takes to arrive, by category: entertainment 3 to 7 days because the value lands on day one, education 7 to 14 days, wellness 7 to 14 days, and health and fitness around 30 days because physical results have to show.] The length is set by how long the value takes to arrive, not by a round number. Entertainment lands on day one; physical results take a month. Finding your own number is less mystical than it sounds. Look at the users who converted and stayed, and find the action that separates them from the ones who didn't. Then look at when that action happens. The trial should end just after that, not at whatever round number your competitor uses. ## 2. Plan structure, briefly Two kinds of people hit your paywall. The ones who already know they'll use this for a year, and the ones who aren't sure yet and are watching the number. Almost every paywall is built entirely for the first group and then treats the second group as a conversion-rate problem. They're not a problem. They're a different customer who needs a different product to buy, and weekly plans are usually that product. That argument has its own piece, because it is bigger than a section: see [monthly subscriptions and LTV:CAC](/notes/monthly-subscriptions-hurt-ltv-cac), which covers how the buyers actually split, which third a paywall can move, and the weekly economics for the ones who will never commit to a year. What matters here is only that trial length and plan structure are two separate decisions and people run them together. The trial answers "how long until they feel it." The plans answer "how do these two buyers each want to pay." Neither answer tells you anything about the other. ## 3. Price level Do the competitive research. Benchmark against comparables. Decide whether you're premium or value-positioned. One thing has shifted: price sensitivity has dropped with AI products, and annual plans that used to sit at $59.99 are holding at $69.99 and $79.99. That said, if you're early and your product is still commoditized, lean lower. A lower price gets you more users, which accelerates learning and lets you iterate faster, which is the whole game at that stage. You raise prices once retention is strong and people genuinely love the thing. ## The takeaway Honest answer: most early-stage products should not be optimizing price at all, and most founders don't want to hear that. They should be optimizing for learning. Offer the trial, and match it to how long your product actually takes to be felt: day one for entertainment, a week or two for wellness and education, a month where a body has to change. The pricing power comes later, and it comes from a product people can't stop using. Most trial lengths I look at were set once, early, by somebody copying a competitor, and nobody has revisited it since. ## Common questions ### How long should a free trial be? Match it to how long it takes a user to experience the value, and set the trial just past that. Entertainment can work at 3 to 7 days, because the value lands on day one. Wellness and education need a week at minimum and often 14 days. Health and fitness is often around 30 days, because physical results have to show up. Find your own number by locating the action that separates users who stayed from users who didn't, then look at when it happens. ### Why do lifetime memberships work for some products and not others? Because they are the far end of the same value-experience question. A lifetime membership works when the thing is something you expect to invest in for the rest of your life: personal investment, wellness, fitness. Nobody buys a lifetime membership to something they expect to be finished with. ### Should I offer annual, monthly, or weekly? That is a separate decision from trial length and it has its own piece: see [monthly subscriptions and LTV:CAC](/notes/monthly-subscriptions-hurt-ltv-cac). Short version: your buyers split into roughly three groups by how much commitment they will accept, a paywall only moves the middle one, and weekly is the plan most teams have not tried for the low-commitment group. ### How should an early-stage app set its price? Benchmark the market, but if you're early and commoditized, lean lower. A lower price buys more users and faster learning. Optimize for learning now and raise prices once retention is strong. --- ## Your team doesn't need to be good at AI. It needs to be top 0.1% Source: https://ottergrowth.com/notes/team-top-ai-power-users Published: 2026-03-16 > I told my team they need to be in the top 0.1% of AI power users. Not top 10%. Not top 1%. Top 0.1%. In practice that means two things: cutting-edge technique rather than better prompting, and automating ruthlessly. The bar we're building toward is 5x the quality of output in a fifth of the time, which is a target we set ourselves, not a measured result. I told my team this week: you need to be in the top 0.1% of AI power users. Not top 10%. Not top 1%. Top 0.1%. Full stop. Here's the uncomfortable part for anyone whose job is to be smart for a living. AI is already more capable than most experts in most domains. As a growth adviser I feel that pressure acutely, and it only gets more intense from here. The people who thrive aren't going to be the most naturally intelligent. They're going to be the best operators of AI. Big winners and losers are being made right now, and I fully intend to be on the right side of that. 🔥 ## "We use AI" stopped meaning anything Everyone says use AI. Fine. But "our team uses AI" now tells you about as much as "our team uses the internet." It's table stakes. It's not a strategy and it isn't a differentiator. Most people use AI like a faster search box. They ask it a question, paste the answer, move on. The top 0.1% are doing something different in kind, not in degree. They rebuild the workflow around it: chaining steps, running agents, building their own small tools, putting verification loops around the output so they can actually trust it. One group saves a few minutes. The other group changes what a single person can ship. [Figure: A distribution of AI skill. The bulk of people use AI as a faster search box. A thin sliver at the far right, the top 0.1 percent, rebuild how they work. The returns live in that tail.] The returns are not spread evenly across the curve. They pile up in the tail, where a few people work in a way most teams have not caught up to. ## So what does "top 0.1%" actually mean in practice Right now, two things. **1. Cutting-edge technique, not just prompting.** Living in tools like Claude Code and Cowork. Running agentic workflows rather than one-shot chats. Using leading prompting practices, and tracking every major model release and capability jump like it's your job. Because it is. The half-life on this stuff is short enough that a technique you learned six months ago is often no longer the good way to do it. **2. Automating ruthlessly.** Your own work. Your team's work. Your clients' work. If you're not building automation into everything you touch, you're building on a foundation that's already cracking. The second one is where most teams stall, and it isn't a skill problem. It's that automating your own work feels like arguing yourself out of a job, so people quietly don't. That instinct is the thing to manage directly, which I'll come back to. ## Why 0.1% and not 1% Because the returns are nonlinear at the tail. The distance between not using AI and using it competently is real, but it's small. A modest speedup. The distance between competent and elite is enormous, and it's enormous for a specific reason: the elite user isn't doing the same task faster. They're doing a bigger task that wasn't possible before. For my own consulting business the goal I've set is aggressive on purpose: 5x the quality of output in a fifth of the time of what someone gets doing it themselves with AI. To be clear about what that number is: it's a bar we set for ourselves and are building toward, not a result we have measured. You don't get near it by being pretty good. You get there by being at the edge, and then by spending the time you save on iteration, specificity and execution confidence instead of coasting on the speedup. That last clause is the part that gets skipped. A team that gets 3x faster and pockets the 3x has bought itself a slightly earlier finish. A team that gets 3x faster and puts it back into more iterations has bought itself better work. ## The reason "get good at AI" doesn't work as an instruction Telling a team to be top 0.1% and leaving it there is a wish, not a plan. I've watched that fail: everyone agrees, nothing changes, because the actual blocker is that nobody has time to figure out which of forty techniques matter this month. So I made the material rather than the exhortation. Everything I've made mandatory for my team, the advanced techniques, the agentic workflows, the whole thing, exists as a package they can work through: a podcast version, a video explainer, slides, and a reference doc. Built with NotebookLM in an afternoon, which is itself the point. There's really no excuse not to be an expert when the curriculum takes less time to produce than one person spends flailing at it alone. That's the honest cost of this standard, by the way. If you set the bar at top 0.1% you owe your team the path to it, and if you don't build the path you're just applying pressure. ## What it looks like on a team Reps, mostly. People building real workflows, sharing what actually worked, and AI becoming the default way work gets done rather than a thing you reach for occasionally. Two things I'd do deliberately. Hire and train for it, so it's a stated expectation and not a nice-to-have someone discovers in month four. And reward the person who automated their own job instead of letting them worry about it, because the alternative is a team that hides its best leverage from you. It's a practice, not a purchase. Buying the tools is the easy part and it's where a lot of teams stop. ## The takeaway "We use AI" is a baseline, not a strategy. The advantage lives out in the tail, where a few people have rebuilt how they work and are shipping at a level the rest can't match. Aim your team there, give it the reps a real practice takes, and hand them the path rather than the pressure. ## Common questions ### What does a top 0.1% AI user actually do differently? They rebuild the workflow around AI instead of using it as a faster search box: chaining steps, running agents, building small tools, and adding verification loops so the output can be trusted. It's a difference in kind, not degree. ### Why aim for 0.1% instead of just being competent with AI? Because the returns are nonlinear. Competent-to-elite is a far bigger jump than not-using-to-competent, since the elite user is doing a bigger task that wasn't possible before, not the same task slightly faster. ### How does a team actually get there? Build the curriculum instead of just setting the expectation. Then reps: real workflows, shared openly, with AI as the default way work gets done. Hire for it, and make sure automating your own job is rewarded rather than quietly punished. --- ## The right go-to-market for a B2C app, from zero to one Source: https://ottergrowth.com/notes/b2c-go-to-market-zero-to-one Published: 2026-03-13 > You do not pick a go-to-market. Your product type already has one, so the work is comparables research plus running the playbook. At zero to one the priority is a product people love, with paid bought as a learning instrument to see where the funnel leaks and to have real users to interview. Organic is the long game. The mistake is chasing it before the product can support it. You do not pick a go-to-market. Your product already has one. Founders treat this as a thing to figure out. It isn't. Most of them get thrashed in different directions by advisors who haven't been in the building game in years. It's simpler than any of them make it sound. The go-to-market for your product type is already known, and the work is finding it and running it: comparables research, then execution. What's left to decide is priority, not sequence. Everything below starts now. The question is where your attention goes when you can only do one thing well this quarter. [Figure: Why the priority is what it is. Paid pays off fast and carries the early months, which is what makes it the channel you can learn from now. Organic stays low, then pays off over 8 to 12 months, and is harder to scale. A product people love sits underneath both and is usually the biggest contributor. This is a picture of payoff timelines, not a running order.] Not a running order. All of it starts now, and it pays off on different timelines, which is why paid is what you can learn from this quarter and organic is not. ## Find out what already works for a product like yours Before anything else, go and look at what works for products like yours. Comparables research, not a workshop. Consumer subscription apps have a known playbook. So do marketplaces, so do tools, so do games. The specifics of your product are yours, but the shape of how it gets in front of people has been run hundreds of times by companies you can go and study. Your job is to run that playbook, not to invent one in a room with your advisors. This is where most of the thrashing comes from. Treating go-to-market as a strategy exercise turns a research question into a debate, and the debate has no answer in it. The answer is already out there in what comparable products do. ## Make something people love. That's the whole job. Everything else on this list exists to get you there faster, and nothing else on it works until this one does. That's why product sits at the top of the priority even though the other work starts at the same time. Most founders underinvest in the two things that actually matter here. The first is whether you are delivering real impact on the user's goal. In health, wellness, and education, if someone does not feel a genuine result, nothing else works. Word of mouth lives or dies right here. You can't out-market a product that doesn't move the user toward what they came for. The second is virality, and not the kind founders reach for first. Referral links and share badges are the bolt-on version, and they mostly do nothing. The version that works is a natural moment where sharing the product creates value for both people. The product itself becomes the reason to share, not an incentive layered on top. That is the whole difference between bolted-on and built-in. [Figure: A bolted-on referral badge is a dead end: the user taps it and nothing loops back. Built-in virality is a loop: sharing the product creates value for both people, which produces the next user and compounds.] A referral badge is a dead end. A product-native share is a loop: each share creates value for both people, and the next user. Here is what built-in looks like in practice. One of my clients makes Pixar-quality stories that kids learn from by talking to AI characters. When a kid gets it, a concept clicks or a story lands, that is a moment parents already talk about with other parents. That word of mouth was happening with or without the app. The product move is to catch that moment inside the product and make it easy to pass along, so the conversation parents were already having gets a digital surface. You are not bolting a referral incentive onto the flow. You are productizing something people already do. If you can't find a native sharing moment, that's a product finding, not a marketing miss. It's telling you something about how people actually use what you built. Fix that before you scale spend. ## Buy traffic so you have something to learn from Paid at zero to one is a learning instrument, not a growth channel. You are buying it to see where the funnel leaks and to have real users worth interviewing. A competent channel on your main ad platform, run properly, tells you which step people fall out of and gives you a supply of people who just made the decision you care about. You cannot fix a funnel you have never watched anyone walk through. Only 5 to 10% of products have a real organic engine at seed or Series A, and nearly every founder believes they're in it. Unless you're one of them, paid is also what gets you off the ground while you build stronger product-market fit and the slower channels. Don't wait for cheap traffic to arrive. You don't win it with better creative anymore. You win it with more creative, faster, fed by a data loop that tells you what's actually working. Volume, velocity, data. That's the game. The team shipping thirty concepts a week and reading the signal beats the team shipping three and arguing about audiences. ## Start organic, and don't wait on it Organic is real, but its payoff usually lands somewhere in the six-month to one-year range, and it's hard to scale at the rate you need while you are still finding product-market fit. Product-led growth tends to be the far bigger contributor. Reddit, SEO, and organic social are all worth building, and you should lay the foundation early. Just don't expect them to carry you while you're still finding fit. Founders fall in love with free traffic like it's going to text them back this quarter. It won't. The mistake almost everyone makes is chasing organic acquisition before the product can support it. Nobody shares a product they don't love, so the channel has nothing to work with. You end up waiting on an engine that needs the very thing you were hoping it would buy you. Chicken and egg, and nobody wins that one. And be careful what you take from other people's success stories. You'll see founders online who scaled to real numbers with no paid spend, and the lesson looks like organic is the play. It isn't. They scaled because the product was so good that people couldn't stop bringing others to it. The non-paid engine everyone wants is what a genuinely great product earns you, not a channel you hack your way into. That's also what investors are actually looking for: a product people love enough to spread on their own, rather than a clever acquisition trick. ## The takeaway Don't start by picking a channel. Start by finding out what already works for a product like yours, then run it. And it's not really an order so much as a priority. Get to a product people love as fast as you can, and buy whatever traffic you need to learn how. Start organic now and expect nothing from it this year. Run it the other way, chasing organic before the product is good enough to keep the users you win, and you spend a year paying to cover a gap no channel can close. ## Common questions ### How do I choose a go-to-market for a B2C app? You mostly don't. Your product type already has a known playbook, so the work is comparables research plus running it. What's left to decide is priority: get to a product people love as fast as you can, and buy whatever traffic you need to learn how. ### Should I start with paid ads or organic growth? Paid, and treat it as a learning instrument rather than a growth channel. You're buying traffic to see where the funnel leaks and to have real users to interview. The common mistake runs the other way: chasing organic before the product can support it, when nobody shares a product they don't love. ### How long until organic pays off? Usually 8 to 12 months, and it's hard to scale early. Lay the foundation now, but don't count on it to carry your first phase of growth. --- ## How to get organic traffic from Reddit, carefully Source: https://ottergrowth.com/notes/reddit-as-seo Published: 2026-03-12 > Reddit is the strictest, most ban-happy platform you'll try to market on, and that's exactly why it works: the community trusts it. Play it as SEO, not community. Reddit threads rank on Google, so you win by owning the threads already ranking for your keywords. The whole addressable universe is maybe 200 to 300 comments, at about 10 a week, and the pacing thresholds are what decide whether it works or gets you banned. Can you actually get organic traffic out of Reddit? Very carefully. Or not at all. Reddit is the strictest, most ban-happy platform you'll ever try to market on. That's exactly why it works. The community trusts it because it hunts down anything that smells like a marketer. Respect that or you're gone before you start. Most people treat Reddit as a community to post into. Wrong game. Play it like SEO. ## The opportunity is search, not karma Here's the thing. Reddit threads rank on Google, and the high-traffic ones in your niche pull thousands of visitors a month out of search. That's the actual opportunity, and it has almost nothing to do with the subreddit itself. You're not trying to go viral. You're trying to win the threads that already rank. Which is a much smaller, much more boring job than "do Reddit marketing," and that's a good thing. It means you can list your targets on a spreadsheet before you write a word. [Figure: A user runs a Google search, a Reddit thread ranks at the top, the reader decides inside the thread, and your genuinely useful recommendation gets the click. Reddit is the SEO surface, not a community to post into.] The whole game is being the recommendation inside a thread that already ranks for your buyer's search. ## The process, in order It's manual and it's slow, and that is the point. Every shortcut in here is also the thing that gets you caught. **1. Find the threads that already rank.** Pull 50 to 100 keyword variations in Semrush or Google Keyword Planner. Search those on Google, not on Reddit. Log every Reddit thread already ranking for them. Those are your targets. Searching on Reddit instead of Google is the most common way I've seen teams start this wrong. Reddit's own search surfaces what's active. Google surfaces what's earning traffic, and traffic is the whole reason you're here. **2. Triage before you act.** Active threads get comments. Archived ones get recreated as new posts. Do the split first, because the two need completely different work and mixing them is how the whole thing stalls. **3. Write like someone who genuinely stumbled onto the product.** Not a pitch. Someone answering the question that was asked, who happens to mention what they use. The writing is the easy part. Finding aged, high-karma accounts to post from is where the juice stops being worth the squeeze for most teams. Expect an extensive Upwork search, and expect that to be the step that kills the project if it's going to die. **4. Every comment gets a supporting comment in the same thread.** Something like "Yeah, I've used this too, solid." It reads organic, because that's how real conversations look. One person recommends, another agrees, nobody writes a paragraph about it. **5. Drip-feed upvotes.** A service like upvote.shop or similar. One upvote every 24 to 48 hours. Beat the comment above you by one or two votes, no more. Over-upvoting is exactly how you get flagged, and it's the impatient step. **6. Track all of it.** Post URL, account, upvote count, flag status. Check weekly. If something gets flagged, that profile goes dark for a few days. **7. Pace yourself.** About 10 comments a week. The whole addressable universe here is maybe 200 to 300 comments. This is not a volume play, and treating it like one gets you caught. ## The four numbers nobody publishes Every write-up of this channel gives you the seven steps. Almost none of them give you the thresholds, and the thresholds are the part that decides whether it works. - **200 to 300 comments.** That's the whole addressable universe, not a starting target. If your plan needs more than that to move your numbers, this is the wrong channel and no amount of execution fixes it. - **About 10 comments a week.** Slower than any growth team wants to run. That's the pace at which this reads as people rather than as a campaign. - **One upvote every 24 to 48 hours.** Not a batch, not a burst on the day you post. The drip is the thing. - **One or two votes above the comment above you.** No more. A comment that jumps ten votes clear of everything around it is the single loudest signal you're not real. Go faster on any of the four and you get caught. That is the entire difference between the teams this works for and the teams that get their accounts burned in a month, and it is why the step that kills the project is usually patience rather than skill. ## Who this is actually for Honest answer: fewer teams than want it. It's worth it for the right product with the right keywords, and it's about as high-intent as traffic gets, because you're meeting people at the exact moment they're deciding. If your buyers genuinely ask "best X for Y" on Google and land on Reddit threads, this is one of the best channels available to you. But it's slow, it's manual, and most teams underestimate how hard it is to not look fake. If nobody owns it week to week, it will not happen. If your product needs a paragraph of explanation before it makes sense, the format fights you. And if you're hoping for volume, this isn't the channel; 200 to 300 comments is the ceiling, not the warm-up. The teams that do well with it treat it like a slow SEO investment that happens to be written by hand, rather than like a campaign with a launch date. ## The takeaway Reddit rewards patience and punishes anything that smells like marketing. Win the threads already ranking on Google, move slowly, look human, and track every account and flag. Done right it compounds like SEO. Done fast, it just gets you banned. ## Common questions ### Is Reddit worth it as a growth channel? For the right product and keywords, yes. Reddit threads rank at the top of Google for high-intent queries, so being the recommendation in the right thread sends you buyers for a long time. It's slow and manual, though, and it's not a volume play. ### Why treat Reddit as SEO instead of community? Because the value is the search traffic those threads already pull from Google, not karma or a following. You're not trying to go viral on Reddit, you're trying to win the threads that already rank. ### How do you avoid getting banned? Move like a real user. Post from aged, high-karma accounts, write comments that sound genuinely human, use a supporting second comment, drip upvotes one at a time every day or two, keep to roughly 10 comments a week, and track flags so a burned account can go quiet. --- ## Six growth questions founders ask, with real answers (when to charge, trials, paid ads, annual) Source: https://ottergrowth.com/notes/six-growth-questions-real-answers Published: 2026-03-10 > Ask an LLM a growth question and you mostly get AI slop and SEO-bait, written by people who have never actually built anything. Here are the six questions I get constantly, with the answers I actually give: charge as soon as possible, set trial length by how long your value takes to arrive, start paid earlier than you think, stop trying to convert the third of your buyers who will never take annual, pricing and product problems usually travel together, and no channel saves a product nobody is excited about. I answer a hell of a lot of growth questions every day. Ask an LLM the same ones and you get AI slop and SEO-bait, written by people who've never actually built anything. Confident, generic, and often wrong in the specific way that costs you a quarter. Here are six I get constantly, with the actual answer. [Figure: What most founders get from an LLM: recycled SEO-bait from people who never shipped. What an operator gives you: answers from having actually built and shipped.] The gap is not intelligence. It's whether the answer came from someone who has shipped the thing. ## "When should we start charging?" As soon as possible. The fastest way to learn what to build is finding out what people will pay for. As Ben Katz put it: if you're not ashamed of what you launched, you launched too late. Charging early isn't about the revenue, and at your volume the revenue is a rounding error anyway. It's about the quality of the signal. Free feedback is polite. People tell you they like it, they'd definitely use it, they'd probably pay for it. A price turns all of that into a yes or a no, and the no is the useful half, because "no" comes with a reason you can go fix. The practical version: put a price on it before you feel ready, and treat the first fifty conversations after that as research rather than revenue. ## "Should we offer a free trial?" Almost always yes. And the length is set by one thing: how long it takes someone to experience the value. Nobody converts on a trial until they have felt the thing working, so the trial has to run at least that long and not much longer. Entertainment can argue the value lands on day one. Somebody watches something and they either liked it or they didn't, which is why 3-day and 7-day trials work there. Education and wellness need a week at minimum, often 14 days. Learning something or building a practice takes more than one sitting to feel like anything at all. Health and fitness needs actual physical results to show up, so 30 days is a lot more common. That isn't generosity. It's just how long it takes to see anything. Then balance that against intent and urgency, because they pull the opposite way: a longer trial gives someone more room to feel the value and more room to forget why they signed up. Products relying heavily on compute are a different animal. There the cost of a generous trial is real money, so a mix of freemium and credit-based tends to make more sense than a time-boxed trial. If you're unsure where you sit, the question is not what your competitor does. It's how many sessions it takes before somebody could honestly say this is working, and then set the trial just past that. The full version of this is in [subscription pricing and trial structure](/notes/subscription-pricing-and-trial-structure). ## "When should we start paid ads?" Earlier than most advisors will tell you. Only 5 to 10% of products have a real organic engine this early. Unless you're one of them, paid is the most predictable lever you have from day one. Predictable is the operative word. The argument against starting paid is usually that it's expensive and it stops when you stop, both true. But the alternative most teams pick is waiting on organic, and waiting on organic is not free either, it just moves the cost somewhere you don't have a dashboard for. Paid buys you a known quantity of users at a known price, which is what you need to learn anything about your funnel at all. Run it while you build stronger PMF and the longer-term channels. Not instead of. ## "How do we drive more annual?" Mostly the wrong question, and this one always lands badly in the room. Your buyers split into roughly three groups by how much commitment they will accept. About a third will never tie themselves to a year, and no paywall you build changes that. About a third take annual for the discount without being asked, and you did not convince them. Only the middle third is actually moved by anything on that screen, and almost all paywall work goes into moving the ends. So push annual at the middle third and stop spending on the other two. What moves that group is mostly layout, and it's cheaper than it sounds: highlight the discount, default the toggle to annual, break the annual price down to a monthly figure so the comparison is obvious, and put monthly and weekly behind a "view all plans" submenu. None of that is a pricing change. Then go do the thing almost nobody does, which is take the low-commitment third seriously. Weekly monetizes them better than monthly, and it is the plan most teams still have not tried. Monthly being low or even negative ROI in B2C is real, and it is why this question gets asked. But the answer is not to fight the third of your buyers who were never going to say yes. It is to sell them something else. The longer version is in [monthly subscriptions and LTV:CAC](/notes/monthly-subscriptions-hurt-ltv-cac). ## "Pricing problem or product problem?" Usually both. Run a PMF read and test price points. Skipping either one leaves you guessing about the half you didn't look at, and in my experience teams almost always skip the same half: they retest price because it's a one-line change, and avoid the PMF read because it might tell them something expensive. Do them together. If the PMF score is weak, no price is right. If the score is strong and revenue still isn't working, now you have a real pricing question rather than a suspicion. ## "How do we get first users without a budget?" Honest answer: most scrappy tactics cost more time than they return, and the ones that actually work are never the ones in the thread you just read. Reddit is nearly impossible unless you know how to game it. Building in public works if you already have an audience, which is a sentence that quietly does all the work in that recommendation. But the real point is upstream of all of it. If the product isn't genuinely exciting yet, no channel saves it. You can pour a distribution strategy into a product nobody talks about, and all you get is a faster read on that. So fix the thing people are supposed to want before you go hunting for distribution. Get the PMF score up or keep iterating until it's there, nail activation, and build basic referral mechanics in early. The product is ultimately your strongest growth lever. ## The takeaway The thread running through all six is that none of them are clever. They're the boring calls you make after shipping enough products to have been wrong a few times, which is exactly what the recycled version can't give you. If you are going to ask an LLM, ask it to reason like an operator who has actually built the thing, and treat any confident, generic answer with suspicion. ## Common questions ### Should founders trust an LLM's growth advice? Treat it with suspicion when it's confident and generic. Most of what an LLM has read about growth is SEO-bait written by people who never shipped anything. Ask it to reason like an operator, and weight the answers that come from someone who has actually built the thing. ### When should a startup start charging users? As soon as possible. Whether people will actually pay is the sharpest signal you can get about what to build, so charging early is a learning decision before it's a revenue one. Free feedback is polite; a price gets you a yes or a no. ### Should a B2C app push annual or monthly plans? Push annual at the third of your buyers who can actually be convinced, and stop spending on the other two thirds. About a third will never commit to a year whatever you do, and about a third take annual without being asked. For the middle group, highlight the discount, make annual the default, break the annual price down to a monthly figure, and put monthly behind a "view all plans" submenu. That's presentation, not a pricing change. For the low-commitment third, test weekly rather than trying to convert them. --- ## 5 lessons from two years advising high-growth subscription apps Source: https://ottergrowth.com/notes/lessons-from-advising-subscription-apps Published: 2026-02-24 > Two years of advising subscription apps across fintech, edtech and health, and the same five patterns show up every time: paid is getting less important and product more, monthly loses money while annual makes it up, signup and paywall are the highest-leverage surfaces, PMF is still unsolved at Series B and C, and when in doubt go back to the problem. Fintech, edtech, health. Same five patterns, every time. I assume all five are true until proven otherwise. Two years of advising subscription apps. Here they are. ## 1. Paid is getting less important. Product is getting more important If people love what you built, they'll talk about it. If they don't, no amount of ad spend saves you. Paid still matters. What's changed is where the leverage sits. The organic engine every founder wants is downstream of a product people can't stop recommending, and you cannot buy your way to that. So when you're deciding whether the next hire is a performance marketer or a product person, the answer has moved. ## 2. Monthly subscriptions lose money. Annual makes up the difference If your paywall isn't actively pushing annual, your unit economics probably don't work, no matter how good your channels are. The mechanism is simple enough. You pay acquisition costs once and monthly hands the user a cancel decision every thirty days, so you're funding a customer who gets to reconsider twelve times a year. Annual absorbs the quiet stretches. This is the one founders push back on most, usually because dropping monthly costs conversion at the paywall. It does. It's still right most of the time. ## 3. Signup and paywall are almost always your highest-leverage surfaces The gap between a great signup flow and a mediocre one is often the gap between a profitable LTV:CAC and never being able to scale. Everything upstream of those two screens is spend, and everything downstream is retention. They're where the money is actually decided, and they're also the two surfaces most teams have touched least recently, because they got built early and then everyone moved on to features. If you only had time to fix two screens, these. ## 4. A lot of companies, even Series B and C, still struggle with product-market fit The pain point isn't strong enough, isn't frequent enough, or both. PMF problems don't disappear with funding. What funding does is let you paper over them for a while, which is worse, because a growth team gets hired to fix what is actually a fit problem and spends a year running competent experiments against a ceiling nobody named. Treating a fit problem as a growth problem is the single most expensive mistake I see, and it's usually made by good teams. ## 5. When in doubt, go back to the problem What specific need are we solving? Who does it actually resonate with? How are we solving it differently? Lost on product strategy? Start there. Not with another tactic. The instinct when growth stalls is to add a lever, and the lever is almost never the issue when nobody at the table can answer those three questions the same way. ## The stage matters as much as the lesson One thing that ties these together: the playbook changes predictably by stage. Pre-launch and launch, then validating your growth channels, then scaling what works. Knowing which stage you're actually in tells you which of these five to weight right now. Lesson 4 is the whole job pre-launch and a distraction at scale. Lesson 3 is the reverse. [Figure: The growth playbook by stage: pre-launch and launch, then growth channel validation, then scaling.] The lessons don't change, but which one matters most depends on the stage you're actually in. ## Why I trust these five None of these are flashy. Not one. They just keep showing up in account after account, whatever the category, whatever the stage. That's the only reason I trust them. That's also why I start from them rather than arriving at them. Walking into a new engagement assuming all five are true is a faster way to be useful than treating each one as an open question, because four of them will hold and the interesting work is finding the one that doesn't. Genuinely grateful to the founders who trusted me with decisions that mattered this much. Best part of the two years by a mile. 🙏 ## Common questions ### What is the biggest shift in consumer growth right now? Paid is getting less important and product more. A product people love drives the word of mouth that ad spend can't manufacture, so the leverage keeps moving toward the product itself. ### What are the highest-leverage things to fix in a subscription app? The signup flow and the paywall. In consumer products they decide the economics, and the gap between a great and a mediocre signup flow can be the difference between a profitable LTV:CAC and not being able to scale. They're also usually the least recently touched screens you own. ### Does product-market fit stop being an issue after raising money? No. Plenty of Series B and C companies still struggle with it, usually because the pain point isn't strong or frequent enough. Funding lets you paper over a fit problem for a while, which is how a growth team ends up running a year of competent experiments against a ceiling nobody named. --- ## \"Do things that don't scale\" is outdated Source: https://ottergrowth.com/notes/do-things-that-dont-scale-outdated Published: 2025-12-04 > Just build it. The trade-off was always the cost to build against the cost of being wrong: building used to be expensive, which made being wrong expensive, which is why you validated by hand first. Build cost is now about a tenth of what it was, so being wrong is cheap. You build it, you test it live in market, the signal is higher confidence because it is a real product, and if it wins you already have it built. Just build it. I hear founders quote the YC line every other week, usually to justify a manual slog. Too bad it's outdated. Not the instinct behind it. The instinct is still right, and I'll come back to it. It's the default that's wrong, because the thing the advice quietly depended on isn't true anymore. ## The equation the advice was built on There is one trade-off underneath all of this, and almost nobody writes it down. **The cost to build it, against the cost of being wrong.** That is the whole decision. Every version of "should we validate this first or just make it" is that ratio, and the reason the answer flipped is that one side of it moved by an order of magnitude while the other stayed where it was. I would go further than "the advice is outdated." I would argue it inverts. ## Why the rule was right: building was expensive, so being wrong was expensive Manual processes genuinely were faster for learning, and the equation is why. Shipping code took weeks. Prototyping meant engineering resources you did not have as a small team, and every hour of eng time spent on a test was an hour not spent on the product. Build cost was high. Which meant the cost of being wrong was high, because being wrong meant burning weeks of the scarcest thing you had on something you then threw away. When the cost of being wrong is that high, you buy insurance against it. That is what the manual work was: a ton of user research and hand-run processes, concierging the service, faking the backend, running it out of a spreadsheet, all so you could avoid paying the build cost until you were sure. You learned the problem while you did it, and you learned it faster than a build would have taught you. That was good advice. It fit its era exactly, because it was the right answer to the equation as it stood. ## What changed: build cost fell by about a tenth, and took the other side with it You can build the real thing now, in an afternoon. With Lovable, Framer, and AI in the loop, I can stand up a real product experience in hours. Not a fake manual stand-in. The actual thing, with a real interface, real flows, and real data moving through it. The cost to build is roughly a tenth of what it was. And because the cost of being wrong was always downstream of the cost to build, it collapsed too. That is the part that actually changes your behavior. Being wrong is now cheap. So you just build it, and you are fine with being wrong, because you test it. That also flips the economics the rule was built on. The scalable path used to be the slow one, which is the entire reason the rule existed. Now it's often the fast one. [Figure: A fake manual workaround is fast to stand up but gives low-fidelity, throwaway signal. Building the real thing with AI and no-code is now also fast, and it gives high-fidelity signal that compounds.] The scalable path used to be the slow one. Now it is often the fast one, and it teaches you far more. ## So the question changed It isn't "manual or scalable" anymore. It's this: what's the fastest route to a learning you'd actually bet on? That second half is the part people skip, and it's the part that does the work. Fast is easy. Fast to a learning you'd act on is the real bar. I've watched teams run a manual test in two days, get a clean-looking result, and then change nothing, because deep down nobody believed the signal. That isn't a fast learning. That's two days. So before you pick the method, say out loud what would have to be true for you to actually move. If the answer is "I'd need to see people use it with me out of the room," a manual workaround can't get you there no matter how quick it is. ## Building the real thing usually wins now Three reasons, roughly in the order that decides it for me. 1. **Higher-fidelity signal.** People react to a real product. They perform for a fake one. When someone knows you're walking them through a mockup, they're being polite and helpful, and polite and helpful is not a buying signal. 2. **It compounds.** You keep iterating on what you built instead of binning it. A manual test ends with a slide. A real build ends with something you can change on Monday. 3. **You're not burning founder weeks on work that evaporates.** Concierge tests are expensive in the one currency you have least of. Your team's time is your scarcest asset. A bit more effort upfront for a learning you'd actually act on is a trade worth taking. And then there is the outcome nobody counts when they run this decision, which is what happens after the test resolves. If it's a winner, you already have it built. You are not writing a spec off the back of a concierge result and starting the build you avoided; you are shipping the thing that just worked. If it isn't a winner, you change it, and changing something that already exists is easy in a way that starting over is not. Both branches come out ahead. That is what "the cost of being wrong is low" actually means in practice: not that you mind being wrong less, but that neither outcome leaves you holding nothing. ## Where this goes wrong Now of course there's a failure mode, and it's the obvious one. "Build the real thing" turns into building for two months. That is not this. The whole argument rests on the build being an afternoon or a few days. If your version of building the real thing is a full sprint, the old math comes straight back and the manual test wins again. What the tools changed is how much you can get standing up in a day, not whether build time costs you anything. So scope it like a test, not like a launch. One flow. The narrowest slice that produces the signal you named a minute ago. If you can't get it standing up in a couple of days, that's information too: either the slice is too big, or this is genuinely one of the cases where manual is still faster. Honest answer: I get this wrong sometimes and end up two days into something I was sure was four hours. ## What's still worth doing by hand Keep the manual effort where the doing is the point. Talking to your users. Closing your first customers yourself. Onboarding the first handful personally. In those cases the manual act isn't standing in for a product, it *is* the thing you're trying to learn, and the relationship you build doing it is half the value. Nobody has automated their way to understanding why someone churned. That half of the YC advice hasn't aged at all. It's the fake-the-product half that has. ## The takeaway Just build it. "Do things that don't scale" was built for a world where building was slow, so being wrong was expensive, so you bought insurance against it by hand. Build cost is about a tenth of what it was, which took the cost of being wrong down with it. Test live in market, trust the signal because it came off a real product, keep the winner because it is already built. Don't default to manual. Default to fast, and fast now usually means building the real thing, scoped small. ## Common questions ### Is "do things that don't scale" still good advice? The instinct is right, the default is outdated, and the reason is one equation: the cost to build against the cost of being wrong. The rule assumed building took weeks, which made being wrong expensive, which made manual validation the cheap insurance. Build cost is now about a tenth of what it was, so being wrong is cheap, so the scalable path is often the fastest one. That is the opposite of what the rule assumed. ### When should you still do the manual, unscalable thing? When the manual act is the point: talking to users, closing early customers, the hands-on work that teaches you the problem or builds the relationship. Also any time a quick manual test genuinely is the fastest route to a learning you'd bet on. Sometimes it still is. ### What should you reach for instead of a manual workaround? Build the real thing with AI and no-code tools, scoped to one flow. You get higher-fidelity signal, the work compounds because you keep iterating on it, and you stop spending scarce founder time on tasks that vanish the moment you finish them. If it can't stand up in a couple of days, cut the slice down or go manual. --- ## Wellness and education app retention is a motivation problem, not a content problem Source: https://ottergrowth.com/notes/retention-is-a-motivation-problem Published: 2025-11-25 > People buy an education or wellness product to feel they are investing in themselves, and that feeling is what carried them through the ad, the onboarding and the trial start. After that, the only thing that sustains them is visible progress toward their own goal. You cannot out-content that. So the fix is to hand progress back at every surface: the ad, the app store listing, onboarding, trial start, the first two days, and every session after. Users don't churn because they ran out of content. They churn because the reason they showed up faded. Nobody buys a learning app for the content. They buy it to feel like they're investing in themselves. That feeling is the entire reason they clicked the ad, read the listing, sat through onboarding and started the trial. It carried them all the way in. Then the learning starts, and learning is hard. Past that first payoff, the only thing that sustains someone is evidence they're getting closer to the goal they came in with. When an education or wellness app has a retention problem, the team almost always reaches for more content. More lessons, more workouts, more guided meditations. It rarely moves retention. Here's the part most teams get wrong: they treat this as a content problem. It isn't. It's a MOTIVATION problem, and the two get fixed in completely different places. Meanwhile every founder is over here going 🤔 "why aren't users as motivated about this as we are?" ## What you're actually competing with It's a fight for attention against Netflix, TikTok, and every other low-effort dopamine hit. It boils down to making perceived value >> perceived effort, every single day, or you lose. That framing matters because it tells you where the lever is. You aren't going to win by making your product more entertaining than TikTok. You win by moving the other side of the inequality: making the value the user perceives from a session obviously larger than the effort it costs them. Most content investment raises neither. ## The motivation is what's leaking Motivation peaks the day someone downloads: a new year, a health scare, a goal, a burst of guilt. Then it decays, fast. If the only thing holding the user is that first burst of motivation, you lose them the week it wears off, no matter how good the library is. The scale of this is easy to underestimate. Only about 25% of people stay committed to a self-improvement goal beyond thirty days. The average person's motivation peaks on Monday and falls roughly 23% by Wednesday. And when you ask people what actually stops them, 67% name the same thing: not a lack of time or money, but a lack of motivation. That is the real opponent, and no amount of content beats it. [Figure: Motivation is high at download and decays. Relying on novelty and new content falls below the action threshold and the user churns. Closing the goal loop, by showing the user the goal they set and their progress against it, holds motivation above the threshold.] The first burst of motivation always decays. What decides retention is whether anything holds it above the line the user needs to keep showing up. ## You cannot out-content a motivation problem You have seen the cycle. Launch the product, focus on content quality, expand the catalog, hit a retention wall somewhere around thirty to forty percent, scramble for solutions, and repeat. Each loop adds more content, and none of it touches the thing that is actually leaking. Piling on content adds choice, not motivation, and choice is not what a demotivated user needs. Sometimes more content makes it worse, because a bigger library is a bigger decision, and a demotivated person does not want a decision. They want a reason to open the app today that does not depend on willpower they no longer have. ## Onboarding should build buy-in, not just familiarity Most onboarding teaches the user where the buttons are. That is familiarity, and it does nothing for the curve above. What you want out of those first few screens is commitment, and there are three ways to get it. Explicit goal setting: ask how many hours a day they want to learn, or how many sessions a week they want to hit, and make them pick a number. Opt-in accountability: ask whether they want a daily reminder, and let them say yes rather than defaulting them into it. And challenge-based commitment: ask what streak they want to hit this week, so the target is theirs and not yours. Most apps do one of the three. The real lift comes from combining all of them, because each one gives you something to reflect back later, and the reflecting back is where retention actually comes from. ## The first 48 hours make or break it If a user has not felt real value inside that window, they are gone, and nothing you send in week three brings them back. So push. Get them through as many lessons or sessions as you can, with roughly three touchpoints a day across email and push. Some users will find that annoying. Do it anyway. The alternative is a polite cadence that respects an inbox belonging to someone who has already churned. ## Close the goal loop Here is the one that moves retention more than anything else on this list, and it has nothing to do with the size of your library. After every session, remind the user of the goal they set and show them how far they have come against it. They told you they wanted four sessions a week. Show them they are at three. They said they wanted to hold a conversation by spring. Show them the ground they have covered since January. Across the education and wellness apps I have worked with, closing that loop is worth something in the range of a 2-3x improvement in retention. Nothing else in this article is close. It works because motivation does not have to be regenerated from scratch every day if the product keeps handing the original reason back to the user with evidence attached. Habit-anchoring helps here as a supporting move. Attach the session to a routine the user already runs without thinking, the morning coffee, the commute, the moment before bed, and the loop has something reliable to run on. One product I work with is Spotify meets Duolingo: you learn a language through the music you were going to put on anyway, so the session rides a habit that already exists instead of demanding a new one. That gets them to the session. Closing the loop is what makes the session worth repeating. ## Progress has to be handed back at every surface, not just in the app Closing the loop after a session is where most of the value is, and it is not the only place the loop exists. The user's goal is what you sold them in the first place, so it has to show up everywhere you talk to them. Not stated once and dropped: reinforced at every surface, so the thing they are working toward is in front of them from the ad all the way through onboarding. The chain, in order: **The ad.** Sell the goal they're buying into, because that is the thing they are actually paying for. **The app store listing.** The value they came for, in the words they would use for it. Someone lands here straight off the ad and decides whether this is the thing that gets them there, so the listing has to be about their goal and not your feature list. **Onboarding.** Ask for the goal, with a number. If you never ask, you have nothing to hand back later, and neither does anything else in the product. **Trial start.** Repeat their number back to them, so the trial is framed as progress toward their goal rather than a countdown to a charge. **Days zero to two.** First evidence, fast. This is the 48-hour window above, and what it is really for is proving the goal is moving. **Lifecycle email, and every session after.** The goal they set, and how far they have come against it. [Image: A post-session screen showing the goal the user set in their own words, their progress against it this week, the ground they have covered since January, and one next step. Underneath, the chain of surfaces where progress has to be handed back: the ad, the app store listing, onboarding, trial start, days zero to two, and every session.] The session screen as something you could build, and underneath it the chain of surfaces the same goal has to travel through. Miss a link in that chain and the product starts to feel like effort with no payoff, which is exactly when people go back to scrolling. ## The takeaway If your education or wellness app is churning, do not start with the content library. Start with motivation. It peaks at download and decays no matter what, so your job is to hold it above the line the user needs to keep coming back. Get explicit buy-in during onboarding, push hard through the first 48 hours, close the goal loop after every session, and hand that progress back at every surface leading in, from the ad onward. Stop trying to out-content a problem that was never about content. Learning will always be harder than scrolling. Always. But when someone can see they're actually progressing, the effort starts to feel worth it. ## Common questions ### Why do education and wellness apps churn even with great content? Because retention is driven by motivation, not library size. The motivation someone feels at download decays, and if nothing holds it up, they leave regardless of how good the content is. ### How do you actually improve retention in a wellness app? Close the goal loop, and close it everywhere rather than only in-app. The same goal runs through the ad, the app store listing, onboarding, trial start, the first two days and every session after. Get an explicit goal, an opt-in reminder and a challenge commitment out of onboarding, push hard through the first 48 hours, and then after every session show the user the goal they set and how far they have come against it. Anchoring the session to an existing daily habit is a good supporting move, because it gets them there without spending willpower. ### Does adding more content improve retention? Rarely. More content adds choice, not motivation, and a demotivated user does not want another decision. It can even hurt by making the app feel like more work. --- ## K-factor math: why referrals won't fix high CAC Source: https://ottergrowth.com/notes/referrals-wont-fix-high-cac Published: 2025-11-21 > A referral program is a percentage of a funnel that already works, which is why it cannot fix a CAC problem. Incentivized referrals add roughly 10 to 15% acquisition and content-based referrals about another 10 to 15%, and neither compounds: below a k-factor of 1 you have a decaying tail rather than a loop. Fix the LTV:CAC ratio directly, and understand that the thing which actually moves the needle is offline, product-driven word of mouth. A referral program is a percentage of a funnel that already works. Which is exactly why it cannot fix your CAC. You are asking it to multiply a number you have already told me is bad. This piece is about that trade specifically. If what you want is how big the referral channel can get and what actually changes your acquisition mix, that is a different question and it has its own piece: [referrals as a growth channel](/notes/referral-growth-channel-framework), which covers the mix, the three levels in depth, and the conditions for word of mouth. What follows here is the unit economics. You have three sources of users. Paid. Offline word of mouth. And incentivized referrals. Incentivized referrals typically add about 10 to 15% acquisition. Content-based referrals add about another 10 to 15%. Both are real and both are worth building. Neither of them compounds, and that is the part almost nobody actually runs the numbers on. Apps with a k-factor above 0.5? Rare, and almost always social or communication tools. ## What the k-factor actually does The k-factor is how many new users each user brings. Those new users bring more, who bring more, and so on. The catch is what happens when k is below 1, which is where almost every app lives. It becomes a decaying series. Start with 100 paid users at a k of 0.15, and you get 15 referred, then 2, then effectively none, and it fizzles out. [Figure: At a k-factor of 0.15, 100 paid users bring 15 referred, then 2, then almost none, a quickly shrinking series that totals about 118 users, an 18 percent lift.] Below a k-factor of 1, each generation shrinks and the whole thing sums to a modest, one-time boost. The math only flips when k gets close to or above 1, where each cohort roughly replaces itself and the loop compounds. That's genuinely rare, and when it exists it's almost always a product-native viral loop, the kind where sharing the product is the product rather than a referral incentive bolted on the side. Most consumer apps sit between a k of 0.1 and 0.2. A useful boost. Not a rescue. ## Build them anyway. Just expect the right thing None of the above means don't build referrals. Incentivized referrals and content sharing are both cheap, both work, and most teams somehow still don't have the first one. Viral features are the only tier that reaches 0.5-0.8 k-factor territory and they only fit specific product types, which is a property your product either has or doesn't rather than a decision you get to make. That is as much as the three levels need saying here, because they are the other post's subject and it does them properly: see [referrals as a growth channel](/notes/referral-growth-channel-framework) for the tiers, what each one is actually for, and how to tell whether level 3 is even available to you. Back to the money. ## Virality lowers blended CAC a little. It does not rewrite it The lift is real and worth building. But it lowers your blended CAC by a slice, and a slice is not a change in the underlying economics. Run it against your own numbers rather than taking it as a slogan. If you are buying users at a CAC that does not clear your LTV, a tail that adds a fifth of your volume moves the blended figure by a fifth of the gap between paid CAC and free. It does not close the gap. It cannot, because the referred users are riding on top of the same funnel, converting at the same rates, monetizing the same way. Everything that was broken upstream is still broken, and now it is broken for slightly more people. If your LTV:CAC is underwater, a decaying referral tail won't float it. You'll just be paying almost as much per user, with a slightly nicer average, having spent a quarter building the thing that produced the slightly nicer average. And this is the part that makes it a trap rather than just a disappointment: a referral program is one of the most satisfying things a growth team can build. It ships, it has a dashboard, the number goes up. It looks exactly like progress. It is also, on a broken funnel, the most expensive way to avoid the actual problem for a quarter. ## Fix the actual problem The real lever is the ratio itself. Raise LTV with better retention, better monetization, and a harder push to annual. Lower CAC with stronger creative, a tighter funnel, and better activation so the users you buy actually stick. Referrals are a supplement to paid acquisition, not a replacement. They improve LTV:CAC and add organic volume, but they're rarely the game-changer people are hoping for. And the thing that genuinely moves the needle here was never a program at all. It's offline, product-driven word of mouth: the product actually being great, plus viral features where the product type honestly supports them. That is the one lever in this whole area with the range to change a business, and no referral incentive substitutes for it. ## The takeaway You can't plan your way out of high CAC with a referral program, because a referral program is a percentage of a funnel that already works. Incentivized referrals add about 10 to 15%. Content-based referrals add about another 10 to 15%. Neither compounds, because below a k-factor of 1 you have a decaying tail and not a loop. Build both, because they're cheap and they work on a machine that's already running. Then go fix the ratio itself, and let referrals do what they're actually good at, which is making a working machine a bit better. ## Common questions ### Can a referral program fix a high customer acquisition cost? Usually not. Unless your k-factor is near 1, referrals add a small decaying tail, roughly a 10 to 20% lift on paid, which doesn't rewrite underwater unit economics. ### What k-factor do you need for real viral growth? Close to or above 1, where each cohort replaces or grows itself and the loop compounds. That's rare, and it comes from product-native sharing rather than a bolted-on referral incentive. Apps above 0.5 are almost always social or communication tools. ### How much do referrals actually add? Incentivized referrals typically add about 10 to 15% acquisition, and content-based referrals about another 10 to 15%. Both are real and both sit on top of a funnel that already has to work. Neither compounds, because below a k-factor of 1 the referred cohort shrinks each round rather than replacing itself. ### What should I build first? Incentivized referrals, then content sharing. Both are cheap and both work. Viral features are the only tier that reaches 0.5-0.8 k-factor territory, and they only fit specific product types, so assess that honestly rather than trying to manufacture it. The tiers are covered properly in [referrals as a growth channel](/notes/referral-growth-channel-framework). ### If not referrals, what actually fixes high CAC? The LTV-to-CAC ratio itself: raise lifetime value through retention, monetization and annual plans, and lower CAC through better creative, funnel and activation. Referrals multiply healthy economics; they don't repair broken ones. --- ## The content wall is dying, and that's good for your app Source: https://ottergrowth.com/notes/the-content-wall-is-dying Published: 2025-05-12 > More content is not more value. A bigger library encourages browsing and kills conversion to actually starting something, because every extra title is another decision and every decision is a chance to close the app. Netflix is moving to fewer options and adaptive recommendations, and B2C content apps, especially education, should follow. Netflix is overhauling its core browsing experience. About time. The content wall is dying. Instead of endless tiles, Netflix is adjusting recommendations as you browse, showing fewer options, and emphasizing previews to lower the barrier to pressing play. That's a bigger deal for B2C content apps than it looks. ## More content stopped meaning more value For years, everyone from MasterClass to Calm copied Netflix's model, assuming content volume equals value. Build a bigger library and users will perceive more worth. There's a fatal flaw in that logic. More content encourages browsing, but it kills conversion to actually STARTING something. Every extra title is another decision, and every decision is a chance to close the app instead of pressing play. [Figure: As a content library grows, the likelihood that a user actually starts something declines, because more choice creates more decision fatigue.] The paradox of the content wall: the bigger you make the library, the less likely a user is to actually begin. ## For education apps this has been a strategic disaster Learners arrive with less clarity on what they want to learn and higher commitment anxiety than someone browsing for a show. Hand them an endless catalog and you've amplified both problems at once. They browse, feel behind, and leave without starting a single lesson. The thing that was supposed to signal value turned into the thing blocking activation. ## The uncomfortable part: the catalog is also your sales pitch Here's the tension nobody wants to name, and it's the reason this is hard rather than obvious. The big library genuinely works, just not where you think. It works at the point of purchase. "Thousands of classes, hundreds of instructors" is a good reason to subscribe, and if you shrink that claim your conversion at the paywall usually suffers. So the catalog sells and the catalog blocks starting, and both are true at once. Which is why the fix isn't deleting content. Nobody is telling you to burn the library. The fix is that the library stops being the interface. You keep everything you have, you keep saying how much of it there is on the pricing page, and then the moment someone is inside, you stop showing it to them. Breadth is a purchase argument. It's a terrible home screen. ## The future is curation, not choice Intelligent curation that removes decision fatigue: surface the right next thing, show fewer options, get people into the product instead of deeper into the menu. In practice that means designing for the first action rather than the fullest catalog. The question for any screen is what the one thing this user should start right now is, and how few decisions stand between them and being in it. A few things that tend to work: - **A default, not a grid.** One recommended next thing, already queued, with browsing available but not the main event. - **Previews over descriptions.** Netflix's own change. Lowering the cost of evaluating something beats writing better copy about it. - **Resume beats recommend.** The strongest next action is usually the thing they already started, and it's the one most apps bury under new releases. TikTok already cracked this. You don't browse TikTok, it starts for you. ## The takeaway Stop treating library size as a proxy for in-app value. More content encourages browsing and suppresses starting, and starting is what activation and retention are built on. Keep the catalog, sell on the catalog, and then curate hard once they're inside: show less, default to something, and design every screen to get the user into the product instead of deeper into the menu. ## Common questions ### Does more content improve a content app? Not past a point, and it depends where you're measuring. A bigger library helps you sell the subscription. Inside the product it adds decision fatigue, which suppresses users starting anything, and starting is what retention is built on. ### Why is the content wall bad for education apps specifically? Because learners arrive with less clarity on what to learn and more commitment anxiety than someone picking a show. An endless catalog amplifies both, so they browse, feel behind, and leave without starting a lesson. ### What should replace the endless library? Not a smaller library. A different interface onto the same one: surface the right next thing, default to it rather than presenting a grid, use previews to lower the cost of choosing, and put resume ahead of recommend. TikTok's "it starts for you" model is the direction. --- ## Monthly vs annual subscriptions: the thirds that decide your LTV:CAC Source: https://ottergrowth.com/notes/monthly-subscriptions-hurt-ltv-cac Published: 2025-01-28 > A third of your buyers will never buy annual. A third take annual without being asked. Only the middle third is moved by anything you build, and that is what a paywall is for. So the job was never converting monthly buyers to annual. It's monetizing low commitment and high commitment differently, and weekly is the plan most teams haven't tried. A third of your users are never buying annual. No paywall you build is going to change that. ## Your buyers split into thirds Here is the read I have landed on after enough of these. Buyers at a subscription paywall split into three groups, and in my experience it is roughly even thirds. A third will never commit for a year. Either they won't tie themselves to anything for twelve months, or they don't trust the plan yet. Some of them will tell you outright. A third takes annual for the discount without being asked. You didn't convince them. They were always going to. The middle third can be convinced. That's the only group your paywall actually moves. [Figure: Buyers at a paywall split into roughly three equal thirds. The left third will never commit for a year or does not trust the plan yet, and is where weekly plans get tested. The middle third can be convinced and is the only group a paywall moves. The right third takes annual for the discount without being asked.] Almost all paywall work goes into moving the two ends. The two ends are exactly the groups that already decided. ## The ends don't move That picture changes what you are optimizing. Most teams run pricing as one job: get monthly buyers onto annual. So the paywall work goes into arguing with the third that was never going to commit, and into re-selling the third that had already made up its mind before the screen loaded. Neither group is listening. One has a reason not to commit that your discount doesn't touch. The other is already yours, and every pixel you spend on them is a pixel you didn't spend on the group in the middle. Run it as two jobs instead. Monetize low commitment and high commitment differently, and stop treating the low-commitment third as a conversion failure. ## Why the middle third is worth the argument The middle third is worth fighting for because of what happens after the sale. Netflix earns its monthly charge because you open it most days. The value is continuous, so the charge always feels like it was worth paying. That's why everyone copied the model, and for anything with a real daily cadence it's still the right call. Learning doesn't work that way, and neither does most of what people buy in order to improve themselves. Someone signs up in a burst of motivation, which is reliably the shortest-lived thing about them, learns hard for a stretch, drifts, and comes back months later carrying a slightly different goal. These are personal-investment products. You buy them against uphill motivation, not toward an obvious pleasure. Nobody needs discipline to open Netflix. Plenty of people need discipline to open the language app they genuinely want to be the kind of person who uses. That distinction predicts the billing problem better than the category label does. Education, health and fintech land here most often, but the test isn't which industry you're in, it's whether using the product costs the user something emotionally before it pays them back. If it does, motivation fades on a schedule, and a monthly plan turns every one of those fades into a live cancel decision. An annual commitment absorbs the fades and gives the outcome enough time to actually arrive. That's the whole mechanism, and it's why the middle third is the group your paywall should be built around. ## Lever one: weekly plans for the low-commitment third This is the lever most teams still haven't pulled, and it's the one I'd test first. A weekly price amortizes to roughly double the monthly one. So even if someone stays a shorter time, you often come out revenue positive on that user compared to the monthly plan they would otherwise have bought. That's the whole argument for weekly, and it's aimed squarely at the third that was never going to commit for a year. They've told you what they're willing to do: pay for a short horizon. A weekly plan sells them exactly that, at a price that reflects what a short horizon actually costs you. It's untested at most companies, which is the other reason to run it. You already know what your monthly plan does. You don't know what weekly does, and finding out is a pricing change rather than a rebuild. ## Lever two: a paywall built for the middle third Everything else goes to the group in the middle, and here the plays are well understood. Lead with annual. Show it broken down to a monthly price so the comparison the buyer is making is the one you want them making. Price the discount loudly. Put monthly and weekly behind a "view all plans" submenu rather than presenting them as the reasonable default. I've written up the plan-structure side of this in more detail in [subscription pricing and trial structure](/notes/subscription-pricing-and-trial-structure). The objection is always the same, and it's fair: does pushing annual this hard cost you conversions at the paywall? Usually a little, yes. Two things about that. The conversions you give up skew heavily toward users who were going to churn in month two anyway, so you're trading a number you report for a number you bank. And that's uncomfortable when conversion rate is the metric your team is graded on, which is the real reason this change is hard to get approved rather than hard to execute. The other half of the answer is lever one. If weekly is in the lineup, the low-commitment buyer you were worried about losing has somewhere to go that's better for both of you. ## The takeaway Stop running pricing as a campaign to convert monthly buyers to annual. Segment by commitment instead. A third has already opted out of a year, a third has already opted in, and the third in the middle is the only one your paywall is talking to. Build the paywall for them, sell the low-commitment third a plan that fits the horizon they will actually agree to, and spend nothing on the group that was always going to buy. ## Common questions ### Should I be converting monthly subscribers to annual plans? That's the wrong job. Roughly a third of buyers will never commit for a year and a third take annual unprompted, so the only group your paywall moves is the middle third. Build for them, and monetize the low-commitment third differently instead of treating them as a conversion you failed to make. ### Why test a weekly plan instead of just pushing annual? Because a weekly price amortizes to roughly double the monthly one. Even if a weekly subscriber stays a shorter time, that user can come out revenue positive against the monthly plan they would otherwise have bought. It's the plan most teams have never tested, and it's aimed at the third that has already told you they won't commit for a year. ### Will pushing annual hurt my conversion rate? Usually a little, at the paywall. The conversions you give up skew heavily toward users who would have churned in month two, so you're trading a reported number for a banked one. Offering weekly alongside it gives the low-commitment buyer somewhere to go. ### Are monthly subscriptions bad for every app? No. Monthly works for daily-use products like Netflix, where the value feels continuous and the charge always feels earned. It struggles on products bought against uphill motivation, where motivation fades on a schedule and every quiet stretch becomes another cancel decision. The useful question isn't whether monthly is bad, it's which third of your buyers a given plan is for. --- ## What \"product-market fit\" actually means, and how to measure it Source: https://ottergrowth.com/notes/what-product-market-fit-actually-means Published: 2025-01-23 > Most startups do not have product-market fit until Series B, and that is what the round is for rather than a verdict on you. So the job through seed and Series A is measuring it simply and then acting on the measurement. PMF is the point where you have real confidence the product solves a genuine need, in a way that can support a profitable business: a Sean Ellis survey for the need, LTV:CAC above 1 for the economics, both and not one. Then run the Superhuman analysis on the result, find who loves the product and exactly why, and double down on that until word of mouth turns up. And the failure I actually see isn't missing PMF, it's having no thesis to test it against, which is what the four-blank exercise is for. Most startups don't have product-market fit until Series B. Not seed. Not Series A. Nobody says it out loud because it sounds like an insult, when it is actually the job description. That is what the money is for. You raise to buy the runway to scale the team and build the product while you go and figure PMF out, and figuring it out stays the main goal right through seed and Series A. Which only works if you can measure it, and most of the conversation about this term is built to avoid measuring anything. Ask five growth people to define PMF and you'll get five different answers, which is exactly why the term survives. It sounds authoritative while committing to nothing. Listen, as a growth advisor who gives high-minded advice for a living, I get the appeal. 🙂 So make them commit. If your investor or advisor can't give you a concrete answer when you press them, take the advice with a grain of salt and watch them squirm. ## A definition you can actually use PMF is the point where you have real confidence the product solves a genuine need, in a way that can support a profitable business. Three load-bearing parts: - **A specific point.** A measurable moment you can say you've crossed, not a feeling that things seem to be going well. - **Solves a real need.** The core problem-to-value hypothesis is validated. People genuinely want this. - **Supports a profitable business.** At minimum, LTV:CAC above 1. Below that you're donating your marketing budget to your users. Both. Not one. That last part is where most definitions quietly cheat. A product people love with economics that never work is not fit, it's an expensive hobby. Growth that looks profitable on a product nobody needs is not fit either, it's a channel that happens to be working for now. [Figure: Product-market fit requires two things to both be true: the product solves a real need, measured with a Sean Ellis survey, and it can be profitable, measured with an LTV to CAC ratio above 1.] Both have to be true. A loved product with broken economics is not PMF, and neither is profitable-looking growth on a product nobody needs. ## How to measure each half For the need, Sean Ellis's survey is still the most reliable read: ask users how disappointed they'd be if they woke up tomorrow and couldn't use the product. The share saying "very disappointed" is your number. For the economics, is LTV:CAC above 1? That's it. If you make more from a customer than you spend acquiring them, the math works and you can go argue about how much better than 1 you want it. Now of course there are a dozen philosophies on this. These two are just the earliest things I can measure with confidence and the simplest to run, which is a different claim than saying they're the most sophisticated. If you already have retention cohorts deep enough to read a flattening curve, read it. Most teams asking me this question don't yet. ## Running the Superhuman analysis on the result The survey number on its own is a reading, not a decision. What turns it into work is the analysis Superhuman built on top of it, and it is four steps rather than a benchmark. **1. Get the number.** Ask the question, and count the share who say they would be very disappointed. Forty percent is the bar people work to, and it comes from Sean Ellis rather than from anyone's intuition. **2. Throw away the rest of the responses.** Not literally, but for this purpose. The people who said somewhat disappointed or not disappointed are not your market yet, and building for them is how a product gets blander. Segment down to the very-disappointed group and treat that group as the product's actual users. **3. Find out what they love, specifically, in their own words.** This is the step that gets skipped, and it is the one that pays. Not "they like the product." The one thing they would fight you to keep. You are looking for a sentence you did not write, repeated by people who have never met each other. **4. Double down on exactly that.** For as long as it takes. Not on the roadmap, not on the thing the somewhat-disappointed group asked for. On the thing your people already love. Then you run it again, because this is continuous rather than a gate you pass. Understand the product. Understand what people love about it. Read the number. Find who loves it and why. Double down. Repeat. **You know you are there when word of mouth turns up.** Not when the survey clears a threshold. When people love the product enough to tell somebody else about it without being asked, because nobody recommends a product that is merely fine. That is the same signal that decides how much paid you have to buy, which is a separate argument in [referrals as a growth channel](/notes/referral-growth-channel-framework), and it is not a coincidence that the two land in the same place. ## The part almost everyone misses: this runs to Series B PMF is stage-dependent, and my read is more aggressive than the one you usually hear. Most startups do not have product-market fit until Series B. Not seed. Not Series A. And that is not a failure, it is the entire point of the money: you raise to scale the team and build the product while you go and figure PMF out, and figuring it out stays the main goal right through both of those rounds. I've sat in a lot of rooms where a founder is quietly ashamed of not having something they were never supposed to have yet, usually because an investor said it like a verdict. If you are at seed and you do not have it, you are on schedule. So the useful question through seed and Series A isn't "do we have PMF." It's "are we set up to find out, and are we actually doing the finding." ## The failure I actually see It isn't missing PMF. It's having no thesis to test it against. A team runs the survey, gets 22% very disappointed, and has no idea what to do with that, because they never wrote down who it was supposed to be disappointing or why. The number is only meaningful against a claim you made in advance. Without one, you're just collecting readings. That's the real gap, and it's fixable in an afternoon. ## The four blanks With early-stage founders I run one exercise first. Four blanks that take an afternoon to argue about and change everything downstream. 1. **Who is the ICP,** specifically enough that you could list ten of them by tomorrow? If you can't name ten actual people or companies, the definition is still a category, not a customer. 2. **What's the hair-on-fire problem** we solve for them, better than anyone else manages to? Hair-on-fire is the bar. Mildly annoying does not get bought. 3. **What alternatives are we actually up against,** including doing nothing? Doing nothing is the one people leave off, and it's usually the market leader. 4. **What two differentiators are we betting on,** the ones that make us two to three times better at that one job? Two, not seven. If everything is a differentiator, nothing is being bet on. Filled in, for a generic B2C habit app, it looks like this. Deliberately ordinary, because the value is in how specific it is rather than how clever. | Blank | Filled in | | --- | --- | | ICP | People who have restarted a running habit at least twice and quit both times, 28 to 45, already paying for one fitness app they barely open | | Hair-on-fire problem | They do not trust themselves to keep going, and every product they have tried made that worse by showing them a broken streak | | Alternatives | A free tracker, a coach at ten times the price, a friend who nags them, and doing nothing, which is what most of them pick | | Two differentiators | Progress measured against the goal they set rather than against a streak, and a plan that survives a missed week without resetting | That is what "specific enough" means. Every one of those is checkable against real users, and every one of them is wrong in a way you would notice. The point isn't to get it right first time. It's to have something specific enough that you can check yourself against it later. ## Then you go and check You build, you learn, and you come back to the four blanks with real usage in hand. Does the ICP still hold, or did the people who actually stuck around look different from the ones you wrote down? Did the differentiation survive contact with real users, or did they like you for a third thing you didn't rank? Was the problem the one you thought it was? That loop is what turns a PMF score from a vanity reading into a decision. The survey tells you where you stand. The thesis tells you what to do about it. ## The takeaway Stop using product-market fit as a vibe you gesture at, and stop treating not having it at seed as a verdict. Most startups don't have it until Series B. Finding it is the job that the seed and the Series A are paying for. So measure it, simply. The need with a Sean Ellis survey, the economics with LTV:CAC above 1, both and not one. Then do the part that turns a reading into work: segment to the people who would be very disappointed, find out what they love in their own words, and double down on that until people love it enough to tell someone else. And write the four blanks down first, because a measurement without a thesis behind it is just a number you argue about. ## Common questions ### What is a simple definition of product-market fit? The point where you have real confidence the product solves a genuine need, in a way that can support a profitable business. It takes validating both the need and the economics, not one of them. ### How do you measure product-market fit? Two measures. A Sean Ellis survey, asking how disappointed users would be if they could no longer use the product, for the need. LTV:CAC above 1 for the economics. They're the earliest things you can measure with confidence, not the most sophisticated. ### Should a seed-stage startup have PMF already? Usually not, and that's fine. My read is that most startups don't have it until Series B, so at seed and Series A you are on schedule without it. Finding it is what those rounds are paying for. The question that matters at that stage is whether you have a specific enough thesis to know when you've found it, and whether you are actually doing the finding rather than assuming it will arrive. ### What is the Superhuman PMF analysis? It is what you do with the survey result rather than another way of taking it. Get the share who would be very disappointed if they lost the product, then segment down to only those people and treat them as your actual market. Find out what they love, specifically and in their own words. Then double down on exactly that instead of on what the less-enthusiastic group asked for. You know it worked when word of mouth turns up, because nobody recommends a product that is merely fine. ### Can you have PMF without profitability? Not by this definition. A product people love with unit economics that never work is not fit, it's an expensive hobby. You need the need and the profitable business. --- ## In growth, your tools quietly decide your velocity Source: https://ottergrowth.com/notes/tools-and-growth-velocity Published: 2025-01-16 > In growth, speed is the game, and a shitty tool is the silent killer of it. The instinct to buy best-in-class for every category is quietly expensive: feedback fragments, syncs multiply, decisions slow down, and none of it appears on an invoice. The answer is the right system, not the best tool per job. Your team isn't slow because it needs better tools. It's slow because it has TOO MANY perfect ones. Tooling is one of the most important topics in growth that almost nobody talks about. Every growth person will tell you the name of the game is speed. A shitty tool is the silent killer of velocity: it slows you down every single day, and you put up with it anyway, on top of the evaluation, contracting, and switching costs. The right tool does the opposite. It lifts team output and compounds as you build momentum. ## Fragmentation is the tax nobody accounts for The instinct is to grab the best-in-class tool for every job. One for chat, one for docs, one for projects, one for whiteboards. It feels responsible. It's quietly expensive. Feedback gets buried across platforms. "Quick syncs" multiply. Simple decisions take three times longer than they should. New hires take longer to ramp. None of that shows up on an invoice. That's the bill for best-in-class, and it never arrives as a bill. 🙂 Which is exactly why it doesn't get fixed. Every line item in the stack is individually defensible, each one was chosen by someone sensible for a good reason, and the cost lives in the gaps between them where nobody owns it. [Figure: A best-in-class stack of many separate tools fragments feedback and adds context-switching cost. The right system, fewer well-chosen tools, compounds velocity.] The future is not the best tool for every job. It's the right system, with as little fragmentation as you can manage. ## The four bars a tool has to clear For early and scale-up stage teams, the right tool in a category usually clears four: 1. **Has the features you'll actually use,** without the bloat you won't. 2. **Reasonable, balanced pricing** for both starting and scaling stages. 3. **Vetted by the industry,** with the resources to still be a front-runner in three years. 4. **Genuinely easy to use,** because ease of use is what drives team-wide adoption. That last one gets ignored constantly. A powerful tool nobody adopts loses to a simple tool everyone actually lives in (every time, not most of the time). Having led growth across a handful of early and scale-stage companies, I've lost count of how many tool evaluations I've run. Consolidation beat "best-in-class" nearly every time. ## When best-in-class is actually right Now of course it isn't always wrong, and I'd be overstating it if I said consolidate everything. The exception is the category that is your actual craft. If experimentation is the thing your growth team does all day, the experimentation platform should be the best one you can run, and you should absorb the fragmentation cost knowingly. Same for analytics if your whole practice is built on it. The general rule is that you can afford best-in-class in the one or two categories where the depth is the work, and you pay for it everywhere else. The failure isn't picking a specialized tool. It's picking four of them, in four categories where a good-enough option inside a system you already have would have done the job. ## How to actually consolidate Two practical notes, because "consolidate" is easy to say and unpleasant to do. **Switching costs are real and they're front-loaded.** A migration costs you weeks of velocity to buy back months of it, and if you run three migrations at once you will simply be slow for a quarter and everyone will remember that rather than the payoff. Do one category at a time, finish it, let people feel the difference. **The decision needs an owner, not a consensus.** Tool choices made by committee optimize for nobody objecting, which reliably produces the bloated stack you're trying to escape. Somebody picks, and the bar they're picking against is the four above. ## The takeaway Treat your tool stack as a velocity decision, because that's what it is. Resist collecting a best-in-class app for every job. Spend that budget in the one or two categories where depth is genuinely your craft, and take the well-fit, easy-to-adopt option everywhere else. You already know which tool your team quietly works around. That's the one to look at. ## Common questions ### Does the tool stack really affect growth? Yes, through velocity. Speed is the core constraint in growth, and a fragmented stack slows every day of work, while the right tools compound output and momentum. The cost is invisible because it never appears on an invoice. ### Is a best-in-class tool for every job the right approach? Usually not. It's right in the one or two categories where the depth is your actual craft, like experimentation for a team that runs experiments all day. Everywhere else, fragmentation costs more than the specialized tool saves. ### How do I choose a growth tool? Four bars: it has the features you'll use without bloat, its pricing is balanced for starting and scaling, it's industry-vetted and likely to stay a front-runner, and it's genuinely easy to use so the team adopts it. The last one is the one people underweight. ### How do I consolidate without wrecking a quarter? One category at a time, finished before the next one starts, with a named owner making the call rather than a committee. Three simultaneous migrations just make you slow, and that's the part the team will remember. --- ## Conversion rate optimization for subscription apps: where to spend your time Source: https://ottergrowth.com/notes/highest-roi-cro-moves Published: 2024-12-30 > CRO is the highest-leverage acquisition lever most teams have and the most underused, and the blocker is almost never technical. It's process. It comes down to two things: the quality of your backlog, which comes from research rather than brainstorming, and how fast you execute against it. Work the levers in this order: data, CTAs, navigation, page psychology, checkout. I ran the CRO program at MasterClass that lifted conversion 25% a quarter, every quarter. You know the shape of the problem it was solving. Conversion stuck at a pathetic 1%, page speed is shit, the CEO keeps asking why, and you have zero engineers on it. It came down to two things, and neither of them was a testing tool. The quality of the backlog, which came from research rather than brainstorming. And how fast we executed against it. CRO is the highest-leverage acquisition lever most teams have and the most underused. The blocker is almost never technical. It's process. Everything below is in the order I would actually run it, and the order is not taste. Data first, because everything after it is a guess without it. CTAs second, because it is the cheapest thing on the list that moves a number. Checkout last, because it only touches people who have already decided. ## The two things, because everything below depends on them Nearly every stuck CRO program I've seen fails on one of these two, and they fail differently. **Bad backlog, good velocity.** The team ships plenty. It's just shipping button colors and hero-image swaps that somebody suggested in a meeting. You get a long list of flat results and a growing belief that CRO doesn't work here. It works fine. You're testing the wrong things because nobody looked at where users actually drop. **Good backlog, bad velocity.** Rarer and more painful. Somebody did the research, the hypotheses are genuinely good, and the team ships one test a quarter because every change needs an engineer who's on something else. A great backlog executed at that speed produces roughly nothing, because you never get enough shots to hit. The first failure is more common. The second is the one that makes people quit. Fix whichever one you have before you touch the tactics below, because the tactics assume both. ## 1. Data analysis first Stop guessing what's wrong with your site. Heatmaps and analytics tell you exactly where visitors drop off. Start every backlog here. Not in a brainstorm. The useful discipline: before writing a single hypothesis, be able to name the step with the biggest drop and roughly how many people it's costing you. If you can't, you're not ready to test, you're ready to look. Most teams skip this because it feels slow, and then spend a quarter testing things that were never the problem. ## 2. Strategic CTA placement and testing Usually the fastest path to a win once you have the data. Place purchase CTAs every one to two screen folds. Keep the primary CTA visible in the hero and nav. Test three to five copy variants at once, not one at a time. That last part matters more than it sounds. Testing variants one at a time is how a good backlog becomes a slow one. If your tool can run a multi-variant test, run it, and spend the calendar you save on the next hypothesis. ## 3. Ruthless navigation simplification Remove low-click nav items. Validate with heatmaps, not opinion. Prioritize the paths users actually take, not your org chart. Every extra option in the nav is a way for someone to wander off the path to conversion. And nav is political in a way the other levers aren't, because each item usually has somebody's team behind it, which is exactly why you want the heatmap doing the arguing rather than you. A/B test before you remove anything permanently. ## 4. Psychology-driven page structure Match the page order to how people actually decide and conversion follows. Hero: clear product statement. Trust: social proof and credentials. Value: emotional drivers and benefits. Action: one clear next step. Get it backwards and no amount of copy testing saves you. This is the one where teams are most often optimizing inside a broken structure, running their fourth headline test on a page that asks for the sale before it has earned any belief. [Figure: A conversion-optimized page structure, top to bottom: Hero with a clear product statement, Trust with social proof, Value with emotional drivers, and Action with a clear next step.] Psychology-driven page structure: say what it is, earn belief, make them want it, then ask. ## 5. Friction-free checkout Break it into small steps. Start with basic info. Build psychological investment as you go. Show progress indicators. Every unnecessary field is a place to abandon (count yours, then cut a third). That last instruction is not a joke. Go count them. Most checkout forms are carrying two or three fields that exist because someone in a different department wanted the data once, and nobody has ever been made to defend them against the conversion cost. ## What to actually do this week The point of a list like this is that you can start it on Monday, so here is the version that fits in a week rather than a quarter. **Monday.** Open your analytics and name the single step with the biggest drop, and roughly how many people a month it costs you. Write both down. If you cannot name it, you are not ready to test yet and the rest of the week is looking, not testing. **Tuesday.** Open your highest-traffic page on a phone and count the screen folds between purchase CTAs. If the answer is more than two anywhere, that is your first test and it is the cheapest one you will run all quarter. **Wednesday.** Write three to five CTA copy variants, not one. Queue them as a single multi-variant test rather than a sequence, because testing them one at a time is how a good backlog becomes a slow one. **Thursday.** Pull the heatmap on your nav and list the items nobody clicks. Do not remove anything yet. Just have the list, so the conversation stops being about whose team owns which link. **Friday.** Count the fields in your checkout, then find the third you could cut. That is one real test live and three hypotheses grounded in data, in a week, without an engineer. Which is the actual argument here: the blocker was never technical. ## The takeaway CRO is a process problem, not a talent problem. Build the backlog from data instead of opinions, then execute it fast enough that the backlog matters. Work the levers in the order they pay: data first, then CTAs for the quick win, then navigation, page psychology, and checkout. If you only do one thing, do Tuesday. ## Common questions ### What is the highest-ROI conversion optimization change? Usually the call to action. Once you have data on where users drop, testing CTA placement and copy is the fastest path to a win: put CTAs every one to two folds, keep the primary one in the hero and nav, and test several variants at once. ### Why do CRO programs fail? Almost always process, not technical skill. Either the backlog came from a brainstorm instead of research, so you're testing the wrong things, or the backlog is good and the team ships one test a quarter, so you never get enough shots. Conversion gains are backlog quality times execution speed, and a zero in either term gives you zero. ### What order should I work on CRO? Data analysis first, then CTA placement, navigation simplification, psychology-driven page structure, and friction-free checkout. Start with the data so every change is aimed at a real drop-off rather than a hunch. --- ## Backlog prioritization: five levels from voting to a bets portfolio (Product Foundations, part 2) Source: https://ottergrowth.com/notes/product-foundations-part-2-prioritization Published: 2021-07-19 _Continuation of [Product Foundations Part 1: Ideation](/notes/product-management-best-practices-for-ideation)_ The job of a Product Manager is to determine a winning strategy to achieving a company goal. They're also responsible for ensuring that the team is working on the most valuable stuff. Strategy might make up 10-15% of the role and execution is the rest. Prioritization is essential to ensuring that the team is doing the highest value work that they can be. Yet oftentimes people don't put in the rigor needed at this crucial step in the process. The job of a PM is sometimes compared to being a "CEO" of that part of the product. And one of the most important jobs of a CEO is to prioritize resources to the most high-priority work. Imagine past companies where you may have felt that the projects that were resourced aren't high-value or you didn't understand how the decisions were made. That wouldn't sit well with you would it? Well it probably doesn't fly to not have a strong prioritization process as a PM either. Read on if you think prioritization is important. ## Preface A few things you'll notice after reading this article: prioritization is less about a person, a process, or expertise in my opinion, and more about mitigating risk. It's less about critical thinking and evaluation and more about letting the users do the prioritization for you. It's less about creating a vision and executing against it, and more about letting the ideal product for your audience unveil itself over time by placing many many smart bets in different investment areas. They say that entrepreneurs are some of the best risk takers, but in reality they are some of the best risk-mitigators. PMs are after all the CEOs of their part of the product, and the toughest first hire at a startup for the CEO is the first product person. Why? Because of the level of responsibility and impact, also brings with it the opportunity for risk. ## The different sophistication levels of prioritization As I was just saying in the preface is that more sophisticated prioritization methods are better at mitigating risk. In fact, the most sophisticated approaches start to look similar to portfolio and hedge-fund strategies at private equity funds. After all, investment funds are some of the smartest, and most sophisticated risk-mitigators in the world. They optimize every investment for the maximum outcomes and return on their investment. Here are the key levels I think about and have seen. It doesn't cover all, but gives you a good way to think about it **Level 1: Intuition** - This is the "I believe this will work" method. It's not hard to see why this approach is prone to risk, well because people are often wrong! But there are some contexts which I'll get into where intuition does make sense. **Level 2: Voting -** Think of it like group intuition where you throw in some measurement. My vote is that voting is more reliable than intuition though. **Level 3: Grading -** Like voting but you've spruced up the votes with more detail than just a simple hash mark. What about the idea warrants a hash mark? Find out by inputting grading factors. **Level 4: Measuring -** With grading you're inputting what YOU think. But how do we know for sure? With measuring you can know FOR SURE. **Level 5: Risk-Mitigating -** Ok so you know for sure what the right prioritization is. Hedging, then, is the practice of overlaying categories or swimlanes on top of that. This allows you to better respond and optimize your investment strategy over the short and long-term time horizons. [![](/blog-images/2021-09-Screen-Shot-2021-09-29-at-12.59.12-PM.png)](/blog-images/2021-09-Screen-Shot-2021-09-29-at-12.59.12-PM.png) ### Intuition - A betting man's sport Like to bet?? Well come on down you're the big winner! I will say that while intuition isn't the highest confidence approach to prioritization, there are scenarios where it is the right strategy. At a base level, PMs need to be applying some level of critical thinking and evaluation of the roadmap ##### When to use intuition If you've been at a company for a while or are relatively senior in the problem-space, you can draw on intuition and achieve more positive and reliable outcomes. The other instance is in early stage companies or products when you're moving fast and don't have enough traffic to do formal a/b testing. Leaning on intuition can help you move quickly, with less confidence, but quick nonetheless. In general, if you need to move fast, sometimes you may sacrifice confidence for speed. But really, getting that extra confidence by doing voting, grading, or looking at comparables doesn't take that much time in the first place. In most situations, however, using intuition as your approach for prioritizing a backlog is not effective and what people end up finding is that well, oftentimes, their intuition is wrong. ##### A humble-man's thoughts on intuition In high-velocity a/b experimentation, on average 25% of tests are winners, 25% are losers, and 50% are neutral or inconclusive. Now, don't get me wrong, I think every single one of our a/b experiments are the bees knees, god's gift to the product, the most epic product improvements I've ever seen! But after enough of your idea-babies are taken out to pasture to be laid to rest, when your chosen feature improvement actually turns out to hurt your metric instead of help it, you learn to not trust your intuition all that much, and you start to understand that you really just don't know in many cases. In fact I'd wager that in ~75% of those cases, I'll be wrong. [![](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.04.39-PM.png)](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.04.39-PM.png) ### Voting - 5 hash marks on epic idea #1 there, Bill Make sure to diagonal hash the 5th one rather than a 5th straight hash mark like a PSYCHO. Voting has an immediate advantage over intuition because of the likelihood that ideas are evaluated and prioritized correctly. Why? because you are relying on multiple perspectives rather than 1. With intuition, there is a single failure point, but when multiple people give input, it helps counteract bias and leverages differing expertise in the group. Many times the grading will align to what a PM would have prioritized in the first place, which is great. However, on average, you find that there could be around ~30% of ideas that either wouldn't have been at the top of the list without the voting, or maybe were graded too high in a PM's mind and after voting they're not in the top quartile. For voting to be successful you should aim to limit the number of votes someone gets (say 5 or 10). Try to make voting anonymous and randomize the options if you can so that items at the top don't get the most votes as people read from top down. A high number of votes overall helps mitigate against some bias where people can choose to use all of their votes or not use all. [![](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.10.32-PM.png)](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.10.32-PM.png) ### Grading Grading is a great way to quantifiably sort a backlog according to how you perceive the cost/benefit of the ideas. The idea behind grading isn't to be 100% accurate, it's to ensure that there is a relative ranking across different factors that are important to you. This controls for bias and you end up with a more confident backlog. In an ideal world you have multiple people input grading on a backlog, which gets the added benefit of checking grades against multiple perspectives. Applying a grading technique becomes more and more important as your backlog size increases, since it becomes difficult to organize and also make decision between two ideas that sound equally good. The most basic grading method I see is Impact divided by effort. Of course, this will give you a backlog that is resource-optimized towards the highest impact. I've talked before about the [ICE method](/notes/impact-confidence-effort-i-c-e-scoring), which is a great and simple approach for grading. Having that last grading category of Confidence is imperative. This is because, as I mentioned in the earlier intuition section, there's a lot of bias that comes along with your own perspectives. Confidence is a measure of how much supporting data, qualitative and/or quantitative, which makes it more likely that the idea will be successful. It's ok to not have any data, and for many ideas you don't. But you should recognize that some ideas you have seen in comparables, or you have heard from users, or has proven successful in a different application. In my post about [ideation](/notes/product-management-best-practices-for-ideation), there are also different idea sources, which can be thought of as having different levels of confidence. [![](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.20.09-PM.png)](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.20.09-PM.png) ### Measuring Grading is all great and fun, but at the end of the day, if you as a human are inputting the grading, then how accurate could it be? Wouldn't it be great if you knew what % of users that the idea would be relevant for, and how valuable it would be for them? Well you can, through polling and measuring or other signals. Sometimes we do a form of measuring of a feature by doing a "hacked" test where it isn't a real feature but it gives us a signal on whether users want it. Another way to measure is to ask users a question with some multiple choice answers they can select from. You can ask them which of these features would be relevant for you, and if solved, how valuable would that be for you? Some people also do a scatter plot to show what the most highly relevant, most valuable features are. We use our measurements and feed them right into our ICE scoring. [![](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.27.03-PM.png)](/blog-images/2021-09-Screen-Shot-2021-09-29-at-1.27.03-PM.png) ### Risk Mitigating Something could score high, but what is the main objective of it and how does it play with the rest of the backlog? Or maybe it scores low, but it should be prioritized to help with the longer-term strategy. This is where risk mitigation comes in, or rather the concept of "making smart investments". One of the most prime examples of risk mitigators are hedge funds. They invest in some areas and balance their investments there with complimentary investments in other areas. This allows them to quickly adjust and maximize returns on their investments as the market changes, as well as minimize their risk in the event that the market changes in a way that would jeopardize their portfolio say if it was all in cruise lines and then a pandemic hits. ##### Portfolio The main idea to take away from this is the idea of a backlog "portfolio". If you have all ideas in a single backlog and rank them against each other, you end up with a strong backlog. However, there are in fact different strategies you can play with that backlog depending on your objectives and tolerance for risk. For instance, some items on the backlog might be high impact, high effort, and low confidence and therefore may not score at the top with ICE scoring, but could still be something you should do - that would typically be a "risky bet". I think of our portfolio in terms of "safe bets", "good shots", and "risky bets". You may want your portfolio to have a distribution of X%, Y%, Z%. Depending on what stage you are in as a company, maybe you make riskier bets because you need a hail mary (hopefully not). Maybe you are somewhere in the middle. Or maybe you have a lot of cash or resources and decide you want to invest in some riskier bets that could push the envelope on innovation. In general though, in product management, I think it's best to take a balanced approach where you aren't doing too much safe or too much that is risky, but still want to make sure you have a little of each, but mostly in the "good bets" category. I generally shoot for 10% safe bets, 70% good shots and 20% risky bets. Sometimes going up on the risky bets side if we need to bust through a local maxima and get us more ground to work with. ##### Swimlanes Another aspect to consider is swimlanes. For us, we separate out our risky bets and actually de-risk them. We do this using "pilot tests" where we will do a hacked test of a product improvement that is either unproven or may lead us down a path that will require significantly more investment as we move forward. By doing a pilot test on the strategy we can get a signal on if it's worth pursuing further. Another swimlane we have is for "opportunistic and team generated". This is because oftentimes ideas that come up, come up for a good reason. In some cases it makes sense to capitalize on a new development. We don't want to have to grade these or hold them off to next quarter planning. Sometimes you need to ensure there is flexibility to move on ideas opportunistically and just see what happens. I would note that this assumes that the ideas are relatively small so you can do this, or at least that you can make a small version of it to test. In our planning, we use measured user problems in an ICE framework at the "themes" level to identify which problems we want to solve. Then we actually do an allocation of number of tests in each theme every quarter according to our betting strategy. ## FAQS #### What if the CEO or an executive wants something prioritized? Oo this is a tough one, but it does happen from time to time and you should expect it will happen more. Sometimes, if an executive wants something, you should just go ahead and do it. If it starts to happen frequently and is a habit, then you have a problem to address and want to make sure you didn't set the wrong expectations. If the request is going to take a lot of effort, resourcing needs to be discussed. In many cases though, if you don't do it right away that's fine. But the key is to ensure that they feel they were heard and that you communicate in detail what will happen and then timeline so that they have the right expectations. Then make sure to follow up with them from time to time so that they aren't wondering where it is in the backlog. #### How do you communicate the prioritization approach to the rest of the company? I think the best approach for this is to do a lunch and learn, or at an all hands. Also every quarter it's good to do a "roadshow" of your quarterly plan and roadmap with relevant departments. You can have a slide on how you prioritized and landed on the roadmap before you share that part with them. #### How often do you do prioritization? We really do it once a quarter, or rather, revisit our prior prioritization. We do shift some things around through the quarter in terms of where we investing our individual bets, but for the most part stick within the themes that we identified. #### Is prioritization different in different company sizes? It can be, since you might want a riskier or less risky strategy depending on where you are. There is also less knowledge and data about users or less developed perspectives, so naturally you are operating with less confidence. #### How do you deal with conflicts in prioritization between different goals? Swimlanes is the best remedy for this. If you are finding that there are competing goals and frictions, it's usually a sign you need a separate swimlane or need to deprioritize what is creating friction. #### How flexible is your prioritization? On the theme level for quarterly planning, not particularly flexible. This is intentional to ensure that we stay focused on the levers that we've identified to have the highest potential. It also is nice that if there are inbound requests that don't fit in our quarterly themes, we have sound reasoning and logic to respond with as to why it's not being prioritized. Within the themes we do allow a lot of flexibility on what hypotheses we test - this reflects the fact that you are always learning and iterating as you go, so your hypothesis prioritization should be able to flex to accommodate that. There is also a learning benefit to staying focused on a particular theme. As you do tests and iterate your learning compiles on the theme as opposed to trying to think about too many user problems at once. Lastly, I employ a dedicated theme each quarter to "opportunistic bets" to allow for ideas that are too good to pass up but don't quite fit in nicely with the quarterly themes. #### How do you do your quarterly roadmapping and planning? After doing user research to identify the top reasons why someone doesn't convert, we do opportunity sizing according to the user problem and then select 2-4 user problems to prioritize for the quarter. Within those "themes" we allocate a certain number of bets or experiments we plan to take based on how many Engineers are on the team and our experiment velocity track record. We also allocate number of bets according to how much we've explored the user problem to date - pilot (a couple exploratory tests), double-down (put as many tests here as possible), and pivot (a couple exploratory tests in a different direction). --- ## Product ideation techniques for a roadmap that works (Product Foundations, part 1) Source: https://ottergrowth.com/notes/product-management-best-practices-for-ideation Published: 2021-02-16 ## 3 keys to ideation success Many people tend to believe that a key responsibility of Product Managers is to come up with ideas. This is true in that you are responsible for having a big, high-quality idea backlog, however, the highest-quality ideation doesn't come from Product Managers. In fact, I try not to come up with any ideas myself if I can help it. Let's start with the fundamentals. There are a few can't-miss aspects of ideation, which, if you just do these alone, will ensure a great idea/hypothesis backlog. They are: 1. Base your ideation topics around user research. 2. Source from multiple, high-confidence idea sources. 3. Ensure a high idea quantity (20+). Let's break these each down, as well as go into some more advanced techniques. ## Setting up foundational user research for roadmap ideation One of my favorite questions to ask in user research is: **"What are your remaining hesitations or questions with doing \[X activity we want you to do\]?"** It's simple, and helps you get right to the point of understanding and ideating around topics that are central to achieving your goal with users. Here's an example of a survey we ran on the MasterClass website: [![](/blog-images/2021-02-Screen-Shot-2021-02-15-at-5.58.03-PM.png)](/blog-images/2021-02-Screen-Shot-2021-02-15-at-5.58.03-PM.png) An example of a user research poll to support idea generation This type of question identifies topics on a "**user problems**" level. User research at this level helps you directionally start thinking about the blockers of the goal outcome for people. I actually use this feedback to prioritize our quarterly roadmap and projects, along with which hesitations we **won't** be addressing. #### The next stage of an idea-oriented user survey The next stage of sophistication with this type of survey is to ask a follow-on question depending on which response they chose. For instance, if they chose "I would only take 1 or 2 of the classes", we would then follow up with a question like "What best describes your interest in our classes?", along with an additional set of options they can choose. These responses are at the "**user solutions**" level. The great thing about going to the user solutions level is that your work is kind of done for you. I said earlier that I try not to come up with ideas myself, and it's true. I would rather that our beautiful users do that work for me and make our life easier. Now even at this level, there are still many ways to approach a solution, and ideas you can generate that can be turned into experiments, but there's much less guessing involved. It can take multiple years of doing this type of research and running experiments until you have a thorough understanding of these hesitation reasons for people and what it takes to overcome them. ## Advanced user research practices #### **Validating user problem addressability and complexity** This refers to the fact that although 60% of people say that X is an issue, you may only be able to solve the issue for a half of that population of users and the other half is not even in your market. Complexity is related to the difficulty, risk, and effort it would take to solve that issue for users. Some problems are more complex and nuanced to solve than others. The most basic way to go about this validation, is to just ideate around these problems, test a lot, and make sure you are reflecting and learning along the way (tied back to the problems). #### **Journey mapping** This is a map of all questions, considerations, and feelings someone goes through in their journey towards your goal outcome for them. It's good to map these questions and considerations to different stages of the journey for a better understanding and stronger hypothesis generation. You generally get this type of info by doing a good amount of qualitative user research, either in-person, or via tools like UserTesting.com. Another source of info can be Google Analytics or Amplitude "top user paths" reports. [![](/blog-images/2021-02-Screen-Shot-2021-02-15-at-6.29.09-PM.png)](/blog-images/2021-02-Screen-Shot-2021-02-15-at-6.29.09-PM.png) An example conversion journey map #### **Creating a conversion equation** This can be any goal that you want a user to "convert to". A conversion equation is a natural follow-on tactic to user journey mapping, in which you look across all of the journey map and identify "in order for a user to get to the goal behavior, these value-driving factors A, B and C, must be perceived to be greater than the costs X, Y and Z for the user". For example, when someone is deciding whether to refer another person to the platform, they are evaluating the referral bonus incentive against the costs of time and reputation implications of referring the platform in many cases. For product engagement goals, they are often evaluating the value props they get from using the product against the time it takes to get that value, and alternatives they could be doing with that time. What is the breakdown of your goal's conversion equation? ## High-confidence idea sources When most people generate ideas for a backlog, what they default to is coming up with ideas themselves or doing a group brainstorm. These are valid tactics, but I would say that they are lower-confidence idea sources because in many cases the ideas aren't supported by a form of data, are inconsistent, and subject to bias. Now don't get me wrong, I definitely do personal and group brainstorming, but it's important to be factoring in at least 1, ideally 2-3, other sources in order to develop a high-quality backlog. Two of my favorites are **direct user research** like discussed in the first section, and **comparables research**, which is essentially going through 5-10 comparable products and taking ideas from what they are doing that you think could apply to your goals. #### Primary idea sources Here are the main idea sources that I think about. I would say that the "confidence level" of the idea sources goes from highest at the top to lowest at the bottom. [![](/blog-images/2021-02-Screen-Shot-2021-02-15-at-6.43.23-PM.png)](/blog-images/2021-02-Screen-Shot-2021-02-15-at-6.43.23-PM.png) Common idea sources There is a blog post I could write about how to take advantage of each of these idea sources, but for now, what's important is just making sure that you are using at least 2 in your process. I would say that comparables research is the easiest high-confidence source to start out with in terms of bang for your buck, but you have to make sure that you are researching "established" products rather than up-and-coming products to ensure that the product approaches have been well proven. #### What goes into an idea being "high-confidence" A high level concept you can apply to any idea source is the idea of a balance between how proven something is or how much data there is behind it, and how relevant it is in the context of your product. A tactic can be very proven at an established company, but the slight differences in product or audience can make that tactic not applicable for your goals. Likewise, if the tactic that you are looking at has direct ties and comparison to your product, but is not very proven, then it is difficult to have confidence in it as well. [![](/blog-images/2021-02-Screen-Shot-2021-02-15-at-7.17.57-PM.png)](/blog-images/2021-02-Screen-Shot-2021-02-15-at-7.17.57-PM.png) The idea confidence components The key though, and the main thing to take-away is just to be thinking of these different elements when evaluating ideas, since it's very easy to make assumptions that cost you valuable resource time. ## Idea quantity There are a few ways to tackle idea quantity, starting with having enough research to go off of, since a lot of ideas can come right from looking through qualitative and quantitative research. Your ideation can be even more fruitful though by practicing these three approaches: #### **Set research quantity goals**. It's that simple. For instance, when I do comparables research of different companies I like to shoot for 10 products to go through, and try to have them be diversified by different goals like some from a particular type of product and others that just have good UX but in a different category. This is naturally going to support a lot of idea generation. The other benefit is that you start to notice trends and similarities between comparables at that scale of research. #### **Refrain from setting high "idea standards"**. I see it all the time, when people frame a brainstorm around "hypotheses", you've unknowingly set a high standard for the brainstorm. For something to be a hypothesis you have to have a thought-through perspective all the way to solution state. You've shut out the possibility for more upstream broad and open-ended brainstorming outputs like just ideas, and user questions. It goes questions -> ideas -> hypotheses in terms of ideation maturity. #### **Set per-person idea goals**. If you're doing a group brainstorm, you want to set an amount of time for the group to come up with ideas, and as a part of that, you need to set a goal to hit for each person. For example, you start the brainstorm and say "for the next 10 minutes let's try to come up with a bunch of ideas around this topic - what's a good number to shoot for, shall we say 10 per person?". There's some psychological studies behind this which you can find online I'm sure, but the fact is that you will get at least 2 times as many ideas from people with this tactic. More sophisticated product teams build these elements into their core processes so that they are always coming up with high quality ideas. You do this by continually doing and sharing user research, noting down ideas and questions from the group outside of dedicated brainstorm sessions, and building an expectation and culture around bringing ideas to the table on an ongoing basis. John Egan, Head of Growth Engineering at Pinterest, talks about this dynamic in his post on "[How to Supercharge Your Growth Engineering Team With Experiment Idea Review](https://jwegan.com/growth-hacking/pinterest-supercharged-growth-team-experiment-idea-review/)" ## Summary Having a strong ideation process may be the most important aspect of a good product development program. Why? Because it is the tip of the spear, it is the top of your product development funnel. Without good ideation, you will have lower-quality features and experiments which will ultimately hurt your ability to meet and surpass your metrics goals. By grounding your ideation in a base of user research, leveraging high-confidence idea sources, and leaning on idea quantity, you set yourself up for maximum levels of impact on the business. --- ## My Experience at MasterClass Source: https://ottergrowth.com/notes/my-experience-at-masterclass Published: 2020-07-03 _DISCLAIMER:_ _The views expressed here are my views only and don't necessarily represent the views of executive leadership at MasterClass or MasterClass in general._ Let me start off by saying that I’ve worked at 10 companies now in my professional career, large and small, with differing cultures and products. MasterClass is the best company I’ve been at, and I’ll tell you why - I’ll tell you why I joined MasterClass and what excites me about where we’re headed, about how I think we stack up in the key areas that I evaluate companies on, what makes MasterClass unique, what the Product org is like, and I’ll sprinkle in some pictures along the way. Cheers! * * * ## Why I joined I joined MasterClass because, well, I believe there is a huge opportunity to improve education for the world and help people be inspired to pursue their passions. As someone who studied Engineering and then didn’t go into that profession and had a winding path of jobs along the way, I have realized just how much traditional education fails us. In fact, most people don’t even like what they do or the company they're at and may never find something that they really love to do.  Is it not an important cause to help people find fulfillment in their life by exposing them, inspiring them, and motivating them to find their passions whether in a work setting or outside of it? Imagine how much happier the world would be, how much more creative, productive, and energized people would be if they were doing things that they loved. That’s what MasterClass does. It’s really quite powerful, and we hear it all the time from our users. * * * ## How MasterClass stacks up ![](/blog-images/2020-07-mc-pickle-1024x636.png) #### The foundational P’s There are 5 of them that I think about, and it’s convenient that they all start with P: - **Product** -  It’s incredibly important to work for a company for which you care about what they are trying to accomplish, because this will directly affect how fulfilled you feel when making an impact there. - **People** - This includes leadership, your manager, your team, and the people at the company overall. People, I think, is always the most important aspect of a company. - **Position** - Depends on what you’re going for, but you should think about is the role scoped correctly and set up for success, are there personal and professional growth opportunities? - **Process** - Has the company figured out good ways to work and streamlined things to make it easy for employees to execute. How organized are they, is their data in order, what tech stack and tools do they use. Such an underrated-important piece of a company. - And **Pay** - Is the pay competitive to other companies that are the same size and in the same industry? Benefits, rewards, all that good stuff. Most importantly, do you feel valued. Compensation isn’t just about $$$. It’s unlikely that a company you’re considering is really strong in all of these, but it’s worth evaluating all.  In fact, I think you’ll find that many companies you consider, work at currently, or have worked at in the past tend to have a less-than-satisfactory score in 1-3 of the areas, which, depending on your priorities at the time, can make for a less-than-great experience for you. MasterClass is one of the few companies I’ve seen that is relatively strong in all. _So how does MasterClass stack up?_ **Product**: I think it goes without saying that the MasterClass product is high quality, well-known, and impactful, especially if you care about education. The fact is that there is no other product in the space quite like it. **People**: The leadership and management overall I would describe as humble, invested, honest and capable. MasterClass has an extremely high bar for who is hired into leadership positions. The teams overall at MasterClass are creative, intelligent, excited, and fun. **Position**: I can’t speak to positions you’re considering, but I can tell you that there is a low turnover rate at MasterClass, people stay a while and are happy with their roles, people get promoted and are able to move between positions if they’d like. These are all signs of great hiring and opportunities. **Process**: Such a relief that MasterClass has a solid tech stack and tools to work off of. I’ve found that we do actively invest in this - we do a fair amount of retros to keep learning and getting better. This is one area that MasterClass wasn’t as strong in in terms of cross-functional processes when I first joined, but has since then closed the gap. **Pay**: Well-run HR organizations benchmark pay and compensation against other companies using industry data to be competitive. MasterClass has a well-run HR org in that sense and pays competitively. There are also some cool perks that we get (no you don’t get to meet every instructor though lol). * * * ## What makes MasterClass unique I do genuinely think that MasterClass is strong in the 5 foundational P’s, which is a big part of why I decided to take the job, but after being there for a bit, let me tell you about what is different at MasterClass than most other companies out there. **Candor**: People talk about being honest and transparent a lot, but not many companies have actually created a culture of candor. An environment that doesn't foster candor feels like this in the workplace: you are in a meeting or just talking with someone and you don’t say what’s on your mind because you’re not sure how it will be received - that’s what I’ve found many cultures to be like. Imagine how liberating it would be to not be afraid to just say what’s on your mind and be honest all the time. That, I think, is very valuable and is the type of environment you want to be in. **Passion**: We’re fortunate that the MasterClass product is so impactful to people and connects with them. We find that the majority of people that apply to positions are fans and subscribers of MasterClass and have a passion for education and our mission already. This creates a really unique environment where everyone is here for the right reasons and the same mission from day 1. In an interview you might end up talking about a MasterClass you’ve been taking, or in the lunch line with coworkers. **Creativity**: And that brings me to the last unique feature of working at MasterClass - creativity. This comes from our shared passion in the product and the nature of the kind of person it attracts. Everyone is in general curious, creative, and intelligent. I have been absolutely blown away by the work that people do and their side projects and the creative competitions we have. MasterClass has a bit of that Netflix dynamic where we have content creation teams mixed with tech teams, it is pretty cool and unique. * * * ![](/blog-images/2020-07-mad-scientists-683x1024.jpg) ## What the product team is like **Teams**: The product teams are set up as cross-functional pods where you have allocated resources to a project from product, design, and engineering typically. This should go without saying, but it is important to note that the resources are fully allocated and not “borrowed” from those different areas. There are 6 Product Managers - Growth Product is oriented by revenue stream and Core Product is oriented to platforms. **Org Structure**: The VP ([Taylor Adams](https://www.linkedin.com/in/tayloradams/)) spans across Growth and Core Product, which is really great and I think is how orgs should be setup in order to ensure consistency and collaboration across acquisition and engagement product teams. Many companies silo acquisition vs. engagement too much, so I appreciate the progressive structure we have on this. Design and our SEO/Publishing orgs also roll-up into Product, which is a great partnership. **Projects**: We decide projects on a quarterly basis for the most part. It’s not uncommon for Product Managers to be doing something totally different one quarter to the next and that’s really cool that we exercise that flexibility and adaptability in order to put our people on the highest impact initiatives. PMs also typically take on multiple projects at a time and while we do stretch ourselves sometimes, the teams are very capable of delivering. **Responsibility**: Product Managers are given a lot of responsibility from day 1 at both senior and junior levels, in part because we are still developing as a company in the grand scheme of things and figuring a lot of stuff out, so we have to give our PMs a good amount of responsibility in order to achieve our goals. **Culture**: I think the most common personality thread in our group is that everyone on the team is extremely intelligent, and in general I would say that our team has a “growth-mindset” along with a strong focus. The reason I mention those things is because it’s very common for some product orgs to be on the other side of that and PMs don't have a growth-mindset, and focus can be all over the place. I think this is a defining aspect of our product team culture and is one of our strengths. And that's it folks! I hope that this was helpful and please leave me a comment or hit me up on [**Linkedin**](https://www.linkedin.com/in/berlinermichael/) if you are interested to learn more. --- ## The Top 5 Reasons Why Companies Fail Source: https://ottergrowth.com/notes/the-top-5-reasons-why-companies-fail Published: 2020-02-18 Over the last decade I’ve been in a number of different roles in Growth and Marketing, across different sizes and types of companies. A recurring theme that I have found which sets the culture and performance of a company apart from others is how leadership sets a shared goal across all departments and is laser-focused on achieving it. In this first of a 4-part blog post I'll look at the reasons why teams are prevented from achieving their growth goals and dig into how 3 of the reasons are actually quite controllable. While it is difficult to overcome challenges w/out the right team or to test into different pricing models, I find that 3 of the top 5 reasons can be remedied with better focus on executing towards a goal. ![](/blog-images/2020-02-Screen-Shot-2020-02-09-at-3.52.35-PM.png) "[The Top 20 Reasons Startups Fail](https://www.cbinsights.com/research/startup-failure-reasons-top/)" - CB Insights Study With better focus on iterating on a product towards growth, teams achieve product market fit much faster, which enables them to get external or internal funding/resourcing and outcompete others in the space. **But what is the goal really?** ![](/blog-images/2020-02-Screen-Shot-2020-02-09-at-3.30.02-PM-675x1024.png) In the book "[The Goal](https://www.amazon.com/Goal-Process-Ongoing-Improvement/dp/0884270610)" by Eliyahu M. Goldratt, famously on the reading list for Amazon execs, Eliyahu conveys a story about a factory in the industrial era, in which, when you look at any individual part, the metrics show that it extremely optimized, but when looking at the throughput of the whole factory it is actually unprofitable and inefficient. The reason behind this is that they aren’t optimizing the factory’s metrics across the system as a whole. Eliyahu argues that the goal of any company simply is to make money, and so all departments of the factory really should be optimizing towards that and any activities that aren’t driving that goal should be discontinued. **The goal = make money $$$** It’s easy to see why this book made the reading list at Amazon, a logistics powerhouse which, even at their scale has few equals when it comes to efficiency and their ability to execute on initiatives that drive revenue. In the next part in this series I cover how to use a [Growth Model](/notes/how-to-use-and-optimize-a-growth-model) to better understand and stress-test your thinking about the optimum resourcing across your growth levers. --- ## 20 Insights From Top Growth Leaders Source: https://ottergrowth.com/notes/20-insights-from-top-growth-leaders Published: 2019-01-28 One of the most impactful, high ROI activities you can do to improve your effectiveness, is speaking with and learning from other leaders in the same area of expertise. In that spirit, I'm sharing top takeaways from my conversations with some of the smartest people that I am fortunate enough to know. Oftentimes, these are things that you can't or wouldn't find on the internet and they give you a different angle of thinking about a familiar topic which then in turn expands your perspective and allows you to achieve greater success. Enjoy! ## Traffic: [**Sujan Patel**](https://sujanpatel.com/) - Managing Director @[Ramp Ventures](https://www.rampventures.com/) @[Web Profits](https://www.webprofits.agency/): “Gaming” or “hacking” Google search with the traditional methods is dead. You can still hack it, but you learn that the only way to really do it is to play by their rules. The key then, is to figure out how to hack growth using their rules and the most scalable, high-velocity approach possible to give Google what it wants. [**Chris Uglietta**](https://www.linkedin.com/in/chris-uglietta-b85a4a1) - Lead Growth Product Manager @[Zynga](https://www.linkedin.com/company/zynga): You can use paid affiliate sources (lower quality), to boost your daily mobile app downloads and achieve top placement in trending and top app lists. Test and calibrate the minimum download volume threshold you need for your category to get top placement and measure the benefit to organic downloads. [**Cecile Lowrey**](https://www.linkedin.com/in/cecilelowrey/) - Growth Manager @[GrubHub](https://www.linkedin.com/company/grubhub-seamless/): Affiliate lead channels for mobile can be littered with bad quality, so there are many companies that aren’t able to get an acceptable LTV / CAC when they analyze the numbers. However, you can significantly optimize these channels by 35%-40%. It takes a rigorous and analytical process, and many times the affiliate partners will not like to give you that level of control, but in the end, you have to force them to in order to get the channel optimization you need. ## Sign-up: [**Chris More**](https://medium.com/@chrismore) - Head of Growth & Services @[Firefox](https://www.linkedin.com/company/mozilla-corporation): You don’t have to make all sign-up flow steps mandatory - using language in the header such as “almost done” along with clear value props can drive desired conversion rates while giving users choice and providing user value delivery further down the funnel. Use a tool like Amplitude to automate the data science of determining the correlation power between specific user actions during onboarding and the desired retention behavior downstream. [**Scott Schmidt**](https://www.linkedin.com/in/scottschmidt12/) Head of Growth & Acquisition @[Stash](https://www.linkedin.com/company/stashfinancial/): Customizing acquisition funnels and flows to your top segments can give you a 20%-30% improvement to acquisition. Determine who are your top 2-3 largest inbound audience segments and identify what their main driver is for signing up - take that and make sure its customized to meet their needs / expectations all the way from paid advertising through the signup flow through on boarding for that segment. Continue to do this on segments where you find large potential interest from at the top of the funnel. [**Chris Woodard**](https://www.linkedin.com/in/cawoodard/) - VP of Marketing @[Tenfold](https://www.linkedin.com/company/tenfold/): Looking at a conversion landing page, we tend to forget that our site visitors are just people, especially in B2B products. When evaluating a conversion landing page, it is important to give it a “real person” test - trying first to understand what the needs are of the target visitor audience, and then whether the page aligns to and is successful with satisfying their needs and questions as a potential customer in a clear and simple way. ## Engagement: [**Nick Allen**](https://www.linkedin.com/in/nicknella/) - Head of Engagement Marketing @[Ipsy](https://www.linkedin.com/company/ipsy/) @[Lyft](https://www.linkedin.com/company/lyft/): Not all sign up friction is a bad thing, if asking more drives deeper investment in the product and loads up future triggers for personalized engagement. Weigh the trade off carefully between optimizing for immediate signup conversion vs. longer term engagement and retention. Put the data you collect to work from day one to show new users the payoff of their time investment as early and often as possible. To drive even deeper engagement beyond 1st time use, consider incentives that push your audience into a repeated behavior. Instead of $10 off the first transaction, try $3 off each of the first 3. **Andrew Watts** - Marketing Manager @[HQ Trivia](https://www.linkedin.com/company/hq-trivia/): You can implement a lot of subtle mechanisms in a product which draw on human psychology to raise their emotion level and build excitement. For instance, if you are going to give someone a prize or really anything they would like to receive, you can have an animation that goes for 3-5 seconds and delays the presentation / reveal of what they desire - thus increasing their excitement and anticipation which in turn amplifies their whole experience with the product. [**Tyler Denk**](https://www.linkedin.com/in/tyler-denk/) - Head of Product & Growth @[Morning Brew](https://www.morningbrew.com/): There are usually at least 2 types of referrers: power and regular. Most early referral programs treat them as the same, but in reality, they are two distinct audiences which require different paths and flows. It can give a high ROI to invest a little time in identifying early segment indicators and making at least 2 tracks from the beginning to fuel higher referral channel acquisition. ## Process: [**John Egan**](https://www.jwegan.com/) - Head of Traffic Growth @[Pinterest](https://www.linkedin.com/company/pinterest/): Empowering people in a growth org to be more autonomous has been one of the biggest drivers of improved speed and effectiveness. Build a culture where every member of the team is expected to act as a mini-PM for their projects, and a culture where every member of the team is expected to contribute ideas of experiments to run. This will help the team move and iterate much more quickly and generate more high quality growth ideas to try out. [**Travis Devitt**](https://www.linkedin.com/in/travisdevitt/) - Head of Growth @[Aceable](https://www.linkedin.com/company/aceable/): There is a powerful, nonlinear compounding effect from experiment wins. Growth teams that can run lots of tests will ultimately get more test wins. Even with a small win on each successful test (+10-25% uplift in key metric), enough small wins added together can compound into dramatic growth rates for the business. This is why testing velocity is so important and why removing constraints for growth teams should be a priority for startup leadership. It's also why small advantages in areas like analytics, audience segmentation, advertising creative/copy, bidding, landing page design, onboarding, and checkout conversion can ultimately feed a virtuous cycle of customer acquisition where gross margin is generated & reinvested in new channel tests, improving customer experience, and other areas. [**Ryan Farley**](https://www.ryanjfarley.com/) - Chief Marketing Officer @[LawnStarter](https://www.lawnstarter.com/): Everything can be executed in a faster, simpler way. Anytime you catch yourself feeling like an initiative you started or are about to start is complex, that is because it is. You should never be thinking in terms of what could be done in a month, or even a week - what could you do in a day, today? Things obviously take more than a day, but you should have thought through and be able to clearly describe what a “1-day” version would be. [**Benedict Dohmen**](https://www.linkedin.com/in/benedict-dohmen/) - Chief Executive Officer @[Benitago Group](https://www.linkedin.com/company/benitago-limited/): Product development doesn’t have to be that difficult and full of uncertain decisions. If you are strong in executing a fast, consumer-centric testing and iteration process which is based on data insights, there is actually a small chance that you don’t deliver a product that people love and adopt. The problem is that for most companies, that is not one of their core competencies. [**Greg Gerla**](http://www.gregorygerla.com/) - Growth Product Manager @[Indeed](https://www.linkedin.com/company/indeed-com/) @[FunnelEnvy](https://www.linkedin.com/company/funnelenvy/): In-person events are one of the best growth channels, and one I recommend for anyone selling to the Enterprise customer. The network effect of having one of your clients speak to prospective clients is more powerful than any webinar. Try hosting a speaker series lunch or dinner where half the attendees are existing and half are new customers. People will be there to network, but as the glue behind the event, your product will come up in conversation. [**Deedee Chiang**](https://www.linkedin.com/in/deedee-chiang-60816940/) - Growth Engineer @[Chime](https://www.linkedin.com/company/chime-card/): Some of the most creative growth ideas come from members outside of the immediate Growth team. Some of the most interesting, untested ideas comes from looking at what companies in outside industries are doing. For example, companies in finTech can leverage and learn from successful tactics pioneered by companies in the movie streaming or transportation space. Startups in the food and service industry can look to what’s worked in the hospitality and banking industry. ![](/blog-images/2019-01-image-1.jpeg) [**Avery Kadison**](https://www.linkedin.com/in/avery/) - Sr. Director of Product, Fraud @[Dosh](https://www.linkedin.com/company/doshapp/): If there are mediums that someone can try to exploit, and it is worth their while, you can be sure that people will try to exploit them. People try to guess at what these “bad actors” will try to do based on known product weaknesses, but the simplest and quickest way is just to observe their behavior and then make hypotheses about how they would change their behavior after you fix the hole. Taking it further, you can run MVP experiments specifically to these groups and see what they actually do, rather than just hypothesizing. ## People & Org: ![](/blog-images/2019-01-image-3.jpeg) [**Cameron Jacox**](https://www.linkedin.com/in/camjacox/) - VP of Growth @[Lark](https://www.lark.com/): If you have a growth mindset, you can usually tell when a hire candidate does right away. They are hungry, data-driven, fast, creative problem solvers. You can trust your gut more than you might think on that if you are in fact looking for those types of people. ![](/blog-images/2019-01-image-2.jpeg) [**Cole Mercer**](https://www.colemercer.com/) - Senior Product Manager @[Bonobos](https://www.linkedin.com/company/bonobos/) @[Soundcloud](https://www.linkedin.com/company/soundcloud/): There are product managers that excel in a number of different areas that make them strong whether that be data analysis, project management, technical aptitude, or experience design among many others. But, one of the rarest and probably hardest to learn skills is being a visionary and good storyteller / presenter. That's something some people have innately but some don't at all - it is crucial to become proficient with selling the vision to internal stakeholder in order to be successful as a product manager. ![](/blog-images/2019-01-image-4.jpeg) [**Scott Boecker**](https://www.linkedin.com/in/scott-boecker-b6975/) - Chief Product Officer @[Dosh](https://www.linkedin.com/company/doshapp/): Everyone is responsible for growth and that has to be the expectation across the company in order to be effective. Everything you do should be able to be tied back to a growth model which ultimately ties back to the financial model so that everyone can see where things stand currently, ongoing improvements, and the plan to achieve KPI targets. ![](/blog-images/2019-01-image.jpeg) [**Zak Cocos**](https://www.linkedin.com/in/zakcocos/) - Growth Product Director @[Indeed](https://www.linkedin.com/company/indeed-com/): There are often two core business goals that are sometimes at odds with each other: growth and quality. Typically growth trumps quality, but if you are thoughtful in separating the two, you can organize your teams and KPIs to address both, delivering a more balanced growth approach that doesn’t sacrifice user experience. --- ## Growth experimentation for early-stage startups, part 3: measure value, not metrics Source: https://ottergrowth.com/notes/growth-experimentation-for-early-stage-startups-pt-3 Published: 2018-08-19 ## 1) You are doing it all wrong! What is your success metric? What is your hypothesis? What are your KPIs? These are inherently flawed questions. Let me tell you why: These types of questions would be relevant if we had done the traditional hypothesis brainstorming methods to come up with our testing backlog. But, as I've pointed out before, having a hypothesis means that you have already come up with a solution, and furthermore, it [biases you](/notes/growth-experimentation-for-early-stage-startups-pt-1) and others towards that hypothesis because you are incented to prove it right. Instead, brainstorming ideas and questions is more innovation-friendly and creates openness to really evaluate and understand the results and then be free to act on them appropriately.       [![](/blog-images/2018-08-value_greater_than_metrics.png)](/blog-images/2018-08-value_greater_than_metrics.png) ## 2) Improve _VALUE_  not **_METRICS_.** Let's play this out. What if we don't have a hypothesis. What if we say: "we have the data and have done the test scoping to believe that this will have a net positive impact on the company". Now imagine how you would analyze the results... You likely aren't imagining whether the hypothesis is right or wrong are you. Likely you are evaluating a lot of metrics holistically, thinking about their impact on the company, and for that matter, thinking about what defines a "net positive impact on the company". And here is the kicker - when you look at the metrics holistically you discover hidden user behavior gems that help you refine your product and generate new ideas that you wouldn't have found if you were focused on specific success metrics. These are the relevant questions you should be asking, because they are the questions that are actually important.       [![](/blog-images/2018-08-mo_moneyyys.png)](/blog-images/2018-08-mo_moneyyys.png) ## 3) False: Mo' money, mo' problems. You may or may not also be thinking that this is a less data-driven approach. But that is where you'd be wrong, it requires you to actually be more data-driven. The key is in defining what a net positive impact to the company is. At the end of the day, and especially if you've read "[The Goal](https://www.amazon.com/Goal-Process-Ongoing-Improvement/dp/0884271951)" by Goldratt, you will understand that the purpose of a company is to make money, or more broadly speaking, to increase the value of the company. You do have a hypothesis, and that hypothesis is always that - this will create significantly more value than it will cost. So, with that in mind, your questions about the test results shouldn't be "what impacts did it have to my success metric?" or "was the hypothesis proven wrong or right?", instead, they should be "what were the positive and negative impacts across all of the key metrics?" and "what would the resulting impact to revenue be if we implemented it?". Also note how this makes the question of what to test and what to implement much simpler. It is not whether it impacts this metric or that one, it is how much can we improve our value at what cost. So when you are evaluating two tests against each other or two product features it is a matter of value at cost.     [![](/blog-images/2018-08-value_lift_equation.png)](/blog-images/2018-08-value_lift_equation.png) ## 4) Don't buy low-value at a high price. Now your choice is straightforward, you wouldn't walk into a store with two items in the same category sitting next to each other and buy one that is priced way higher than the other now would you. Why don't people do this? I'm not sure to be honest, but my best guess is 1) That it requires effort and is not the easiest path because you actually have to understand and map the relationships your KPIs have with revenue and 2) People have a hard time quantifying what is not easily quantifiable, say "time on the app" - it's not that it's not quantifiable in terms of revenue, it is just that you haven't spent the time and brainpower to figure out how to quantify it. What is that classic manager saying again? "Don't bring me excuses, bring me answers!"? Something like that. So if you have quantified your tests and results in terms of revenue, the go-forward decision is straightforward: will this make us more money than it costs us? Otherwise, it obviously wouldn't make sense. And here we are talking about engineering hours and any other associated costs. The second piece is whether it makes _enough_ revenue at that cost that it is worth our while. For this, I generally defer to the 5% aggregate minimum impact to a core KPI rule of thumb for early stage, but it really varies by company and stage.   The go-forward / no go-forward decision is largely situation-specific. --- ## Growth experimentation for early-stage startups, part 2: scope the test, then prioritize Source: https://ottergrowth.com/notes/growth-experimentation-for-early-stage-startups-pt-2 Published: 2018-06-19 ## Playing a game of blind darts. Ok so you have a bunch of ideas that you think will help grow the company, inspire user engagement, reduce friction, or impact whatever it is that your company growth goals might be. How do you determine what you should test and validate first? In an ideal world you can actually calculate the estimated impact to KPIs along with the estimated engineering effort it would take to implement the solution should the test be successful, and sort rank from there. Unfortunately, that is not always the case, but the same approach still holds true. Remember [the equation that we are trying to optimize for](/notes/growth-experimentation-for-early-stage-startups-pt-1) at a high level: achieving the most value, in the shortest time, as simply as possible. In this stage (test scoping & prioritization), we are optimizing our process for time:   [![](/blog-images/2018-06-value-over-time.png)](/blog-images/2018-06-value-over-time.png) ## Have a destination. The first thing that people tend to miss when determining what to test is considering what solution would be implemented if the test was successful, and what level of effort would come along with implementing that solution. What about any risks as well? It's easy to think of ideas without really reflecting on what the long-term implications might be. Thinking from multiple angles and time-horizons on a single subject takes brain horsepower and effort. If you test without considering this, you are flying in the dark as to what type of results and confidence level you would need to make a go / no-go decision on implementing it as well. What if what you are testing doesn't have a real long-term sustainable solution that you could implement EVEN if the test was successful? If you haven't thought through that, well, then you are in some shit. I think of it like this: [![](/blog-images/2018-06-working-back-from-the-end.png)](/blog-images/2018-06-working-back-from-the-end.png)       ## Then pick the most efficient path to get there. So what do we need to consider anyway? If I came to a product manager or the engineering team and said, "hey we should implement this" without having thought about what that entails, damn would I feel stupid. It is your job in growth to have thought through these things and tee it up for other departments to execute on. I will talk about how we evaluate and prioritize ideas before testing them, but first, let me dispel the myth that all testing is done by A/B or multivariate experiments. Not true. In fact, if your product is in its early stages (less than ~2 years since launch) and you are doing A/B testing of a feature or acquisition channel, you have probably skipped a few stages. Recognize that A/B testing is typically an approach for optimizing something that is already proven, so by definition, if you are A/B testing a product feature you are making the assumption that the existing solution is proven, when in most cases within the first 2 years, it's not. The end goal is to determine how we can achieve the growth goals in the shortest time, so we need to get answers that help us decide how to get there. There are many ways to get answers, and furthermore, certain approaches are better suited or faster to get to answers than others depending on what answers and confidence we are seeking. Here is a framework I think about: [![](/blog-images/2018-06-question-and-test-framework.png)](/blog-images/2018-06-question-and-test-framework.png) When a company is in its early stages, many of the questions are strategic in nature, about where should we start, where is there an opportunity. As you progress and iterate, you learn where there are opportunities, what you should do to tackle them, and lastly how you should do it to achieve the maximum results. I've laid out 5 of the most common ways that I use to get to answers roughly ordered by granularity, but it really depends on your specific use case. One of the most worthwhile exercises before settling on a test method is to try to see if you can walk upwards on this funnel. Similar to the benefit behind the "[5 Whys](/notes/testing-principles-not-solutions)" method, we often make assumptions that aren't proven, so by challenging ourselves to validate those higher assumptions or less granular tests, you naturally simplify your approach and end up gaining additional speed. It will have been time well spent, trust me.   ## Possibly one of the best uses of time: evaluating prioritization. We have an estimated impact to company growth goals, we know how we plan to test it, how do we evaluate where our efforts will be best spent from there? You might have heard of the classic "I.C.E." scoring method, which stands for **I**mpact, **C**onfidence, and **E**ase. The thing is, the ICE method falls short because it is just considering the testing side of the equation and not the implementation. I propose a new acronym: ICESRR. **I**mpact, **C**onfidence, **E**ase, **S**peed, **R**esources, and **R**isks. [![](/blog-images/2018-06-ICESRR-1.png)](/blog-images/2018-06-ICESRR-1.png)   You have impact to company growth KPIs. The ease (the antonym of difficulty) of executing a test. Speed is how long you estimate it will take to reach results on the test. Resources is how many resource-hours it would take to implement if successful. Risks covers any potential downsides or unknown effects that the solution might have if implemented. And lastly, confidence is how sure you are in the scores you gave. Additionally, we give each scoring factor a different weighting - how important is each to your org and decision making? It varies depending on your culture and by stage of development of the company. Then you make a calculation multiplying each score by their weighting, adding them together to come out with a combined "value score" which gives you a way to fairly compare and sort rank tests against each other and determine which would be most valuable to prioritize first. Ideally, the scoring of these factors is derived from hard data and analysis. For instance, the impact score is calculated from an estimation of impact to a growth goal using the exposed [audience size multiplied by the estimated KPI lift](https://growthhackers.com/articles/growth-experimentation-for-early-stage-startups-pt-1/) as discussed in the last post. The effort is determined by an engineering hours estimate of the solution that would be implemented. When your growth process becomes more developed, you should have the KPIs and their associated impact on revenue modeled out. With some basic assumptions and estimations you can boil this down to a simple ratio of revenue / engineering hours.   ## Think differently, think faster. In the next post, I will discuss how you can execute scoped tests quickly, and thought processes for keeping the implementation of successful test _simple._ Remember from the [last post about test ideation](/notes/growth-experimentation-for-early-stage-startups-pt-1), during test scoping and prioritization, the primary objective we are optimizing for is speed. And so I leave you with this thought: [![](/blog-images/2018-06-10x-speed-question.png)](/blog-images/2018-06-10x-speed-question.png) --- ## Growth experimentation for early-stage startups, part 1: growth is a process Source: https://ottergrowth.com/notes/growth-experimentation-for-early-stage-startups-pt-1 Published: 2018-05-29 ## "Growth" is a process, not a magic potion. The idea of having a "Growth" org for startups has evolved over time. Initially called "Growth Hacking" because of the combination of skills from marketing and coding. And also because of famous "hacks" like the classics from Dropbox and Airbnb. Unfortunately, the golden era of finding magical hidden gems that will propel a company towards its goals are over. Although, a tip of the cap to Bird and the other scooter startups recently for my new favorite hack where they just decided to [blast their scooters all over the place -](https://www.cnet.com/news/the-electric-scooter-invasion-is-underway-bird-ceo-travis-vanderzanden-leads-the-charge/)"do first, ask forgiveness later", love it. In today's world, Growth is a structured process. You don't just throw a bunch of shit at a wall and hope a golden ticket to Willy Wonka's factory ends up sticking there now do you? You prioritize ideas, test them, implement those that are successful, rinse and repeat. 3 phases: Ideation -> Testing -> Implementation. In each, you want to optimize for either value, time, or simplicity:   [![](/blog-images/2018-05-ideation-testing-implementation-1.png)](/blog-images/2018-05-ideation-testing-implementation-1.png)   ## What is "the goal"? If you don't get this right to begin with, then you are already F'ed. The goal of a Growth team or really any team in a company is to increase the value of the company. A difference from other departments is that increasing the value of the company is the only goal and is the key purpose of a Growth team. Not just one part of the company, but the company as a whole, and of course, as fast as freakin' possible. I think of it like this:   [![](/blog-images/2018-05-value-over-time-2.png)](/blog-images/2018-05-value-over-time-2.png)   You want to optimize your value over time and do it in the simplest way possible. It's kind of like a value realization rate.     ## The "5% Rule". The majority of companies define what dictates the value of the company in the form of yearly company goals. These are after all, what will determine the valuation of the company in VC's eyes if you have defined them correctly. It is important when brainstorming to think about a specific company goal because it will force you to focus in and you develop more mature theories as you think longer on a single subject. Also, it is more difficult for your brain to brainstorm without a specific category to think about vs. focusing in on a specific subject which creates natural synergies in the thinking process. There are a couple factors to keep in mind when brainstorming that, on the most basic level, determine the potential of an idea. They are impact to a metric, and audience size. [![](/blog-images/2018-05-impact-and-audience-size.png)](/blog-images/2018-05-impact-and-audience-size.png) If you don't think it will be able to change a metric significantly, or doesn't reach a large enough audience, chances are it won't be a high-potential idea. A good baseline target to think about during early stage testing is a minimum of a 5% potential impact to a company goal. Lower than this and it isn't worth testing, and you can probably think of other ideas that can.     ## Questions are more accessible than hypotheses. Some people will tell you to think of ideas or hypothesis, but this basically means that you have already thought things through and have formed a perspective. Yet, in most cases, you or people in the company haven't done this, so it makes it more difficult for the brain to come up with something. I find it's easiest to brainstorm in the form of questions, this leaves it open ended and is more aligned to the actual source of the ideas. For instance: "do our users need more education on how to use the product?". You can think of questions basically like hypothesis, where we believe the default answer to the question is "yes". Yes we think that our users need more education to use the product.   Where do we source these questions? Questions can come from a number of different places, here are some of the main ones: [![](/blog-images/2018-05-ideation-sources-1.png)](/blog-images/2018-05-ideation-sources-1.png)   During the ideation phase, you can employ [the "5 whys" method](/notes/testing-principles-not-solutions) that I posted about earlier to help zero in on what the core principle is that you are trying to test. Also, by doing the simple estimations discussed in this article such as the potential impact to a KPI, and the exposed audience size, you can determine which questions have a high value potential if we were to answer them.   In the next post I will go through grading and prioritizing ideas as well as how to simplify down tests as a way to increase your speed. --- ## Testing Principles, Not Solutions Source: https://ottergrowth.com/notes/testing-principles-not-solutions Published: 2018-05-03 [![](/blog-images/2018-05-efficiency-v-accuracy.png)](/blog-images/2018-05-efficiency-v-accuracy.png) We all have a bias when we think up solutions and here is why: Efficiency over accuracy. When a piece of information is presented to your brain, you better believe it is trying to make sense of it in the quickest and easiest way possible. Most likely dating back to a primitive caveman survival instinct that some great psychologists have published numerous articles on, but that I am too lazy to research right now. See? Efficiency over accuracy.     ## Lazy comparisons. How do we do this you ask? Well, we simply compare everything against principles and experiences we have learned in the past as the fastest path to understanding it. We are all lazy, and it is much easier to compare than it is to think from a blank slate. The brain receives a new problem it must solve, brain says “what have I seen before that relates to this?”, brain applies principles learned from those experiences to decide the next action. But were those principles universally true? True when applied to the current situation? Did you make any assumptions in the process? It's tough to know if you don't have a process for knowing. Adding to this, we always seek, whether consciously or not, to confirm those principles and experiences that we have had. Because what if, god forbid, we found that we were wrong about those experiences or beliefs we have based actions throughout our life on, how would we cope with that kind of reality? Sounds like a hassle if you ask me, I’d rather not. Just look at me confirming my own beliefs in this very post, is this a self-fulfilling prophecy or what.     [![](/blog-images/2018-05-principles-v-solutions.png)](/blog-images/2018-05-principles-v-solutions.png) ## Test principles, not solutions. Now that we know that everyone processes information in an inherently biased way (including ourselves), we can move on to the next phase, which is reverse-engineering what your brain ended up with, and by doing so, you can determine the principles and assumptions you made to get there. Here is an example: You believe that a better onboarding email sequence will be effective in increasing app revisit rate, that is a solution. A principle behind that solution might be something like "people need education before engaging with the product". Which do you think is easier to test? One requires a whole sequence to be built out which might not end up working if our execution misses the mark. One is a yes or no question with the freedom to test it in a variety of ways and save us a ton of time. What if we hadn't read this amazing article and instead of proving or disproving the principle we tested the solution, it failed, but we felt a different version of the solution would work and we tried that and it failed.. and again..well you get the point - sooner or later, we learn that the principle is in fact not true, and hopefully the org is reflective enough to learn that sooner than the later.. let's hope. That would be an example of the principle not being true, but it took us many more and longer cycles to determine that same thing. See, the key to speed is making your tests simple, and solutions are not as simple to test as principles.     [![](/blog-images/2018-05-imageedit_6_7103696182.png)](/blog-images/2018-05-imageedit_6_7103696182.png) ## Imagine yourself as a 5-year old. Asking that constantly curious question of "why?". Why? Why? Why?... One of the best and quickest ways to determine the principles behind a solution or idea is to employ the “5 whys” method. The concept is that a “why?” question is a backtracking type of question, and by asking yourself that 5 consecutive times makes you walk through your logic and naturally brings the discussion to concepts and not specifics. "Why" brings the discussion upwards in terms of granularity. Another way to think about it would be "what would have to be true about human nature, behavior, or the product to make my solution make sense?". In the case of the onboarding email sequence example earlier, users would have to not be familiar or comfortable with your product for that type of sequence to make sense. See, the same result as before, but I would rather determine if they need to understand the product better before I went ahead and tested an onboarding sequence. Now that we know why we believe a particular solution would be successful, and what principle we want to test, the next step involves how you can determine the quickest way to test it. In the next post, I will discuss some thought processes you can apply to help make the test you come up with simpler and quicker. --- ## Minimum Viable Value (MVV) is the new MVP Source: https://ottergrowth.com/notes/mvv-is-the-new-mvp Published: 2018-03-22 ## Minimum Viable What? If you’ve ever worked at a tech company you have probably heard the term MVP, or “Minimum Viable Product”. If you haven’t, you may be thinking that I am sports illiterate, but MVP basically means a first pass at making or doing something. Here is my beef with “MVP” - it is a contradiction - a **V**iable **P**roduct is not a **M**inimum solution to a problem to begin with, it's a product. To build a product or do something you should have some confidence and reason to do it. You wouldn't invest a million dollars and spend a bunch of time making say an automatic laundry-folder unless you were pretty damn sure people were going to use it now would you? Maybe you would, I mean hell.. I would help you make that, it sounds great.   ## C is for "Confidence" It's a tradeoff - how much confidence do you need to be able to make a decision? It really just depends on how much potential payoff you think you could get relative to the effort it would take to figure it out. That's that ROI. At the end of the day, what you are trying to validate with a MVP is the **VALUE** to current or potential users right? So by thinking that there needs to be a product, you have already over-solutioned. In fact, there are many examples in the past few years  of entrepreneurs that have very successfully disproven the need for a product at all, take for instance most kickstarter campaigns or the 500k person Tesla Model 3 preorder waitlist. Today you don’t need a product to sell a product, and you certainly don’t need one to validate the value or even the potential impact of an idea.   ## Act As If Ah yes, the classic self-help motto "act as if" or "fake it 'till you make it'. But here is the twist, you actually CAN just fake it. Make stuff up, fake it, don't be dishonest or anything, or do but don't get caught, doesn't matter to me. But if you have to cut some corners to figure out where to invest your time and resources, DO IT. In Silicon Valley it is the wild west of stuff that doesn't actually exist currently or even come to fruition. Kickstarter campaigns aside, you can simply throw up a nice looking landing page along with a product idea and get funding with no actual MVP product. There are countless examples of products that promise that they are running some "complex matching algorithm" and literally they actually are just outsourcing to some people they hired in the Philippines to do some manual process that mimics the results. Because, well, who cares? We can make the algorithm later.   ## Why is This Difficult for Humans? Guys, let's give credit where credit is due, it is no one’s fault that we over-complicate our solutions to problems. We do it ALL the time! We are after all, only humans; we think, we analyze, we have opinions and all sorts of prior experiences that shape how we make decisions. The problem with the way our brains are conditioned to think is that we always practice newly learned concepts by applying them to real-world, tangible problems. And by doing so we anchor ourselves, and limit our brain's thinking capabilities to what is real and known. So it follows, that it is very difficult for us to think about "fake" scenarios, things that aren't real and might never be. We ought to be doing more thinking in the abstract space when we are problem-solving. More thinking about the overall principles of the problem we are trying to solve, and that will actually help you approach it through a simpler frame. When we think about specifics, well, things get specific, and that is just complex and too much effort.   ## What Would Elon Musk Do? A question I find myself pondering for days on end, as you no doubt do as well. Elon is what you call a "Visionary Entrepreneur", he thinks based on principles, not real situations. When you are testing an idea or trying to prove if something can or can't be done, the key is to remember that you are in fact "testing" it. If the way in which you are testing something is at all close to resembling what you would want to do long term, well then you didn't test it, you in fact just did it. People often ask me "how would we do that though?" when I am testing an idea that is difficult to tie back to the current "real" situation. However, this is the wrong question to ask. The question to ask is "what COULD we do?". Notice the difference. In the first question you are rushing to the conclusion that we are going to take a specific action - you see, you have already solutioned before we have even gotten results! When we get results, the next step is to ask what we could do to achieve those results given how motivated or unmotivated we are by them. The difference is that we are proving out principles, not a specific solution. When you free your mind to think in this way, you open up creativity and the ability to innovate in alignment with the natural problem solving process.   In my next post I will talk about the thinking process to go through when approaching a problem. You can identify the underlying principle you are seeking to understand so that you can focus in your testing on it. --- ## 7 Keys To Innovative Ideas From The Book \"Originals\" By Adam Grant Source: https://ottergrowth.com/notes/7-keys-to-innovative-ideas Published: 2017-11-13 **The book "[Originals: How Non-Conformists Move The World](https://www.amazon.com/Originals-How-Non-Conformists-Move-World/dp/0525429565)" is chock-full with lessons on original thinking and implementing change. Written by Adam Grant, an Organizational Psychologist at the Wharton School of the University of Pennsylvania, you'll find that he draws liberally on insights from psychology studies and famous success stories to make a compelling case for readers, similar to what you might find in a Malcom Gladwell book. It is applicable for small tech companies to established giants, and especially entrepreneurs or those seeking to bring about new improvements or ideas to an organization.** **I've written a short summary sentence for each theme along with some of my favorite quotes and takeaways, enjoy!**   ## [![](/blog-images/2017-11-compassicon.png)](/blog-images/2017-11-compassicon.png) * * * ## Thinking of and recognizing original ideas The way we evaluate our own ideas as well as others, and the causes of false positives and negatives. - People's opinions of themselves or their ideas are always subject to confirmation bias. People rate their ideas and probability of success 40% higher than that of their peers. - Seinfeld was initially rejected by all industry execs but one, and people never thought that buying glasses online would fly until Warby Parker proved otherwise. - People that downloaded Chrome on their computer rather than accepting the default browser performed significantly better on the job - they didn’t accept the default program and proactively chose to challenge the default. - Entrepreneurs are significantly more risk averse than the general population. - Larry page and Sergey Brin almost didn't start Google because they were worried about their futures in PhD research, they initially tried to sell Google for less than $2 million. - A creative mindset is 50% more accurate than a managerial one at identifying successful ideas. - Involvement in the arts such as writing, acting, dancing, and magic makes you 12-22x more likely to end up winning a Nobel prize.     ## [![](/blog-images/2017-11-bellicon.png)](/blog-images/2017-11-bellicon.png) * * * ## Why timing is a driver of success How procrastination is often used as an important tool for idea development, and a market's readiness for change. - People who procrastinate and give their mind time to engage in divergent thinking, produce 30% more creative results. - MLK never had "I have a dream" in his written speech, he winged it starting at that point. - The problems we try to solve are not that complex when you boil it down, but because we typically don't take the time to do so, our solutions usually are. - First movers are six times more likely to fail than followers. - Three out of four startups fail because of premature scaling - making investments that the market isn't yet ready to support.     ## [![](/blog-images/2017-11-thumbsupicon.png)](/blog-images/2017-11-thumbsupicon.png) * * * ## Keys to identifying and forming alliances What makes us more or less likely to form an alliance with another person or group, and how to think about your relationships. - The more strongly you identify with an extreme group, the harder you seek to differentiate yourself from more moderate groups that threaten your values. - Shared execution tactics are an important predictor of alliances, even if they care about different causes. - For insiders, the key representative of the group is the person who is most central and connected, however, for outsiders, the person who has the most extreme views is perceived as the representative. - Ambivalent relationships are unhealthier than negative ones. It takes more emotional energy and coping resources to deal with people that are inconsistent.     ## [![](/blog-images/2017-11-eyeicon.png)](/blog-images/2017-11-eyeicon.png) * * * ## How personal identity plays into divergent thinking Things that make us inherently more likely to engage in original thinking, and why we take significantly more action on something if our reputation is at stake. - Younger brothers were 10 times more likely than their older siblings to attempt to steal a base. Laterborns tend to have a higher propensity to take risks because of their upbringing. - People are much more likely to take an action when that action is tied to their identity. For instance: "don't cheat" vs. "don't be a cheater". - It's much easier to link our agendas to familiar values that people already hold. Instead of assuming that others share our principles, or trying to convince them to adopt ours, we should present our values as a means of pursuing theirs.     ## [![](/blog-images/2017-11-groupicon.png)](/blog-images/2017-11-groupicon.png) * * * ## Fostering originality in groups Examples of how openness to feedback (or lack thereof), caused the downfall of some companies like Polaroid, or conversely, is a core strength of the Bridgewater hedge fund. - The commitment hiring blueprint has the lowest failure rate for startups, followed by the star blueprint and then the professional blueprint. The commitment, star, and professional blueprints make up 14 percent, 9 percent, and 31 percent of hiring. - Social bonds don't cause groupthink, overconfidence and reputational concerns do. - Bridgewater reverses the standard of voicing opinions in your job - in most organizations people only give negative feedback beyond closed doors, and are often punished for raising dissent. At Bridgewater they're evaluated on whether they speak up, and can be fired for failing to challenge the status quo. - Prioritizing values and principles is important because the relative importance of multiple values guides action.     ## [![](/blog-images/2017-11-hearticon.png)](/blog-images/2017-11-hearticon.png) * * * ## The role of emotions as a change agent Converting debilitating emotions to action enablers, and framing change in a way that agrees with our natural inclination to avoid risk. - Reframing anxiety as excitement increases results by 50%. - When we're anxious, the unknown is more terrifying than the negative. - The easiest way to encourage non-conformity is to introduce a single dissenter. - When options are framed as gains, 4 out of 5 people choose the safe option that lets us hold onto and protect what we have. Conversely, if options are framed by what they will lose, we're willing to do whatever it takes to avoid that loss, and 4 out of 5 people will choose the riskier option, taking a gamble that it will significantly reduce our losses. - Superb presentations start by establishing what is: here's the status quo. Then, they compare that to what could be, making that gap as big as possible. - When we're angry at others we aim for retaliation or revenge. But when we're angry for others, we seek out justice and a better system. You can also listen to the podcast with the author from Andreessen Horowitz at [https://a16z.com/2016/02/02/originals/](https://a16z.com/2016/02/02/originals/) --- ## The 1 Step That Everyone Forgets w/ Decision Making Source: https://ottergrowth.com/notes/the-1-step-that-everyone-forgets-w-hypothesis-testing Published: 2017-05-11 I was leading a brand awareness initiative to [build LawnStarter Little Libraries for homes across Austin](https://www.lawnstarter.com/blog/texas/austin-tx/little-free-library-austin/); In only a week I had thrown together a [landing page](http://book.lawnstarter.com/LittleFreeLibraries) and a few email lists I scraped. Now it was time to send out an initial email and measure conversions. Then, in a 1 on 1 meeting with Ryan, the COO of LawnStarter and my manager at the time, he asked me how many people had signed up so far - to which I explained that none had yet and we were about to send out an email. Right in that meeting we made a simple Google form and [posted it to a Facebook group](https://www.facebook.com/groups/LawnStarterLittleLibrariesAustin/); it got an amazing response from the group along with many sign-ups - so we had validated in 30 minutes that there was some demand when we didn’t have much concrete to go off of before, but what if I had spent a week setting everything up for a proper test and it flopped? That’s when I realized that the most commonly overlooked step in hypothesis testing is validation; how can I be confident that the time I am investing in doing a test is likely to be fruitful? [![](/blog-images/2017-05-LFL-Facebook-Post-1.jpg)](/blog-images/2017-05-LFL-Facebook-Post-1.jpg)   ## The challenge with testing, is time. You see it again and again in startups - a backlog of hypotheses that are in queue to be tested. In some cases they are the most high-value items to be done and in that order, but many times it can be done much simpler and quicker. Just like a product manager manages the backlog and product roadmap, so too does marketing need to be smart and deliberate with prioritizing initiatives. After all, time is the most important asset to a startup is it not?   ## Don't forget the basics. Keep in mind that there are defined phases to solving any problem, and here they are - the step that everyone neglects? Validate: [![](/blog-images/2017-05-problem-solving-steps.jpg)](/blog-images/2017-05-problem-solving-steps.jpg) The thing is, many people are too confident in and rely too heavily on their existing subject expertise, and think they have some tests already in mind that are clearly the most high-value, but what discovery and insights are those based off, are they truly customer-centric, and how do you know they are likely to succeed - did you validate?   ## You only hurt yourself if you don't start with discovery. If you haven’t gotten feedback on your hypothesis, haven’t talked with customers, haven’t compared it to other hypotheses, chances are that it isn’t fully developed or won’t give you the best ROI for your time. Typically, the biggest mistake made in this step is not talking with the customer. Sure, you can analyze some datasets, talk with other subject experts in the company, do some critical thinking - but while you are at it, call a customer or 3 or send out a google form to a small audience, it’s not that hard! Yet this is probably the most important stage, and it is where many veer off path from customer-centric to brand-centric solutions. There are two audiences that you need to cover on any question: those that are your biggest advocates and those that were close to converting but didn’t, don't operate in a bubble!   ## Analysis doesn't have to be that complex. Here is a simple approach that I use for prioritizing the insights and ideas that come from discovery in the first stage. Everyone has their own approach, but then again.. _do they_? Don’t shoot from the hip, do something like this: [![](/blog-images/2017-05-Hypothesis-Sort-Ranking.jpg)](/blog-images/2017-05-Hypothesis-Sort-Ranking.jpg) Ease = How easy to execute. Impact = Anticipated results to the bottom line or KPI. Feasibility = How confident you are that it will be successful. The other grading factor I would think about adding is speed - how fast you can determine the potential ROI of the idea.   ## MVP your MVP by validating. Ok so you have your hypothesis and have graded it to determine that you believe it to be the most valuable, and you probably have some idea of how you would like to test it. Now take the core insight that drives the idea and simplify it down to its core, for example, with the LawnStarter Little Libraries it was that people wanted these Little Libraries so much that they would sign-up to host one in their front yard - how would you test that? If It takes more than a day or two to validate, that is too long, hell it should take only a half day if you do it right. Take whatever you were going to do and simplify it down, and then simplify it down again. And I know what you are thinking, all of this takes time, I thought that is what we are trying to optimize for!  - Yes, that is true, but what if you spend a week or two on a test that flops, then look back at how much time was wasted? You don’t have to validate just one hypothesis at a time, you should be able to validate the top 5 on the list in a day or two, then you have the confidence to commit one to two weeks of time, and what’s equally important is getting that additional layer of feedback and insight from validation that helps you refine and polish.   ## Go forth and conquer. Don’t skip over crucial steps in the problem solving process, you will thank yourself when you realize you made smarter decisions and have a more customer-centric outcome. Have a method, VC-funded startups can't afford to prioritize like it is the wild west anymore. And do more by doing less, you are doing too much, just do what you are doing, but less. ---