App Growth Annual 2026: Five Growth Experiments Worth Testing
A practical recap of the main-stage ideas from App Growth Annual 2026, translated into five testable experiments for onboarding, paywalls, win-backs and AI pricing. The recurring theme was to segment users, match the offer to real usage, and judge early conversion by downstream value.
The three ideas that kept recurring
Three patterns ran through the day. First, averages hide useful differences: the same flow can work for one segment and fail for another. Second, the monetization model should follow how people actually use the product, including temporary needs and expensive AI usage. Third, an initial purchase is an incomplete scorecard; refunds, cancellations, return visits and cost to serve change the conclusion.
Babbel: break a rule only when the trade-off is understood
Babbel's contrarian history included moving to premium in 2009, which filtered for users willing to commit to language study and funded the business, while narrowing the audience and raising marketing pressure. Longer onboarding about study time, reasons for learning and prior experience only paid off because the product could act on those answers. Web checkout brought data, speed and fee advantages, but the team also had to keep investing in the full web-to-app journey. Lifetime plans suited a habit people dip in and out of, yet they end the monetization path unless new products or hybrid offers follow. Win-back work therefore grew alongside acquisition.
Burner: design for temporary needs and returning users
Burner found that strong install-to-paid conversion coexisted with weak first-month retention because many people needed a number temporarily. It converted installs to paid at almost 6x the median, but M1 retention was 38%, below the median. More than a quarter of people who cancelled came back within a year — 2.6x the utility app median — with zero effort from Burner; returners made up about 10% of MRR and over 30% of new subscription revenue. Instead of blocking departure, the app offered a $1.99 pause plan, versus $9.99 standard, to hold a number, saving about 10% of would-be cancellations, and tested letting numbers stay active briefly after expiry so users could see what they would miss. Returners skipped standard onboarding, saw what was new, received a more generous trial and were led toward annual pricing, giving them 2x the trial conversion of new users and 5x the annual take rate. Overall, a returning user's subscription was now worth 25% more than a new user's, and returning-user LTV was up by 35%.
Tolan: price AI tiers around willingness to pay and cost to serve
Tolan's voice-first AI product had grown to more than one million users and 100k+ subscribers, with 70+ million AI conversations, showing how inference costs change growth math: more conversations can mean more cost before more margin. The team first attracted a younger audience that installed but rarely paid, then rebuilt its creator programme around older users with everyday use cases and ability to pay. A hard paywall protected margin but capped reach, so the model moved to freemium with a personalized paywall during onboarding and feature gates later. Lite, Plus and a $200/month Unlimited plan were set around cost to serve, with expensive capabilities starting high and moving down only as costs fell. Subscriptions were combined with top-ups, with ads and a family plan under test, and monthly plans were judged on conversion, retention, billing success and revenue per user rather than assuming annual is always best.
Ravi Mehta: build confidence, not just shorter flows
Ravi Mehta reframed onboarding length as a confidence question: what does a person need to feel sure at the decision point? In one example, a three-step checkout for a weight-loss programme converted below target and produced unsustainable refunds, while replacing it with a 25-step intake covering medical history, lifestyle, goals and motivation lifted conversion by 40% and refunds went way down. Useful friction asks relevant, answerable questions; unhelpful friction demands effort before value or asks something hard to answer, so wording matters. The weakest flows sit in a compromise middle, neither thorough nor lean. A stakes-and-complexity view helps decide: high-stakes, complex products need more confidence-building, while low-stakes, simple products should remove everything non-essential.
Paywall roast: the paywall and lifecycle messages are one journey
A live paywall critique compared pairs of popular apps and followed the paywall into permissions, push, email and in-app messages. The repeated lessons were to demonstrate value before asking for payment, state trial terms plainly, request notification permission in context with a clear benefit, and keep relevant contact with people who cancel a trial. The connection is practical: a paywall that is hard to understand undermines trust, while lifecycle messages that arrive at the right moment can help a user reach the value the paywall promised.
Other highlights and market context
Beyond the main stage, the event included 36 workshops, a quiz-style game show on subscription data, and an after-party performance. The game show answers, drawn from the State of Subscription Apps report, were: 1) 55%, 2) 4.6%, 3) 14,700, 4) 17–32 days, 5) “Continue”, 6) Progress bars, 7) Asia-Pacific (APAC), 8) 41%. Market context shared during the product keynote was that 107,000 apps have gone live so far this year, up from 41,000 in all of 2025, with all-time annual revenue tracked now at $17 billion+; per Appfigures, app launches are up 85% year over year while consumer spending grew just 15%; and as a cohort, apps launched in 2023 and after grew 124%, while apps launched before 2023 declined -2%. The implication drawn on stage was that software alone is a weaker moat when building is easier; brand, content, network effects and customer understanding matter more.
| Experiment | Change to test | Measure |
|---|---|---|
| Onboarding commitment | Add one or two goal or commitment questions | Completion, trial-to-paid, early activation by channel |
| Returner experience | Welcome-back paywall or tailored offer for cancellers who return | Conversion, plan mix, lifetime value |
| AI cost tiers | Add a tier or top-up where usage and cost vary | Contribution margin, conversion, retention |
| Confidence audit | Remove or rewrite one doubt-creating onboarding step | Trial conversion, refunds, early cancellations |
| Trial clarity | State what users get, when billing starts and how to cancel | Trial starts, trial-to-paid, early cancellations |
Each experiment should be run against the current flow and segmented where channel or user type changes the result.
Frequently asked questions
When does longer onboarding help instead of hurting conversion?
Babbel's example was that extra onboarding questions helped when they reinforced commitment and the product then delivered the personalization promised. The test is whether completion, trial-to-paid conversion and early activation improve by channel, rather than whether the flow is simply shorter.
How should apps treat returning subscribers differently from new users?
Burner's experience was that people who cancelled often returned on their own, and returners responded better to a shorter welcome-back path, a more generous trial and annual pricing presented as best value. The practical first step is to measure how many cancellers return and what new subscription revenue they create before building a separate experience.
Why does hybrid monetization fit AI apps with high compute costs?
For an AI app, growth can raise serving costs as fast as revenue. A hybrid approach can keep a lower entry point while charging higher tiers or one-off top-ups for expensive capabilities, with tiers reviewed as serving costs change. Segment users by usage, revenue and estimated cost to serve before adding a tier.
How did Who Wants to Be a MillionARR? work and who won?
The qualifying round was open to the whole room: 10 timed multiple-choice questions answered from attendees' phones, won by whoever answered fastest. The top three — Hamad, Daphne, and Adrian — then took to the stage for a Jeopardy-style final with real buzzers and categories covering the app market, paywalls, retention, and AI apps. Prizes were $500 for first, $300 for second and $200 for third, plus a championship belt. Daphne Tideman won after dropping to -800 points and then answering question after question correctly.