Maxim Borisik leads product and experimentation efforts at Adapty, a platform that helps subscription apps run monetisation experiments at scale, from paywall A/B testing to AI-driven revenue optimisation. His background spans ad tech and API-driven SaaS products, giving him a cross-industry view of how growth strategies translate into measurable results.
We sat down with him to talk about how AI is reshaping the way subscription businesses decide what to test, what to trust, and what to build next.
1. When experimentation moves from a single app to thousands running in parallel, the way most teams approach testing has to change. Based on your experience, what breaks first: the infrastructure, the statistics, or the team’s process?
Maxim Borisik:
Everyone expects it to be infrastructure or stats. It’s neither. Those are the parts you can engineer around, and at platform scale that’s our job, not the customer’s. What breaks first is the human layer: teams stop trusting the results and quietly stop using the harder features. Going from one app to thousands, the number of decisions grows faster than any team can confidently interpret. So the real failure isn’t “can we run the test,” it’s “do we believe the answer, and do we know what to do next?” You can see it in the funnel: roughly 80% of teams reach a recommendation, and only a small fraction ever launch it. Nothing technical stopped them. The infrastructure keeps working long after the team has checked out, and that’s the thing you actually have to design for.
2. One of the bigger shifts in mobile monetisation has been the ability to change paywalls without waiting for app store releases. You’ve made that possible at scale. How much does that test velocity change the way teams think about experimentation?
Maxim: Less than I expected, with one clear exception. The mature teams, the ones already experimenting as a habit, are genuinely in love with it, and that’s where the real gain shows up. That’s exactly who our Flow Builder is for: they can A/B test entire paywalls and change them without shipping an app update, so the release cycle no longer rate-limits their learning. But that’s the point. Velocity raises the ceiling for teams that already know what they’re doing. It doesn’t create a testing habit where there wasn’t one. We gave plenty of teams the ability to change a paywall in minutes, and they still never ran a single experiment. For them, the technical block just turns into decision paralysis: “I can change anything now, so what do I change?” Speed multiplies an existing habit; it doesn’t start one.
3. It’s still common for teams to judge a paywall test by early clicks or Day-0 conversion rather than what it earns over time. Why do you think so many subscription apps still optimise for those signals instead of real revenue or LTV, and what does that cost them?
Maxim: Because early signals are fast and legible, and real LTV is slow and noisy. You get Day-0 conversion the same afternoon. A true 12-month read takes about a year. So people optimise the number they can show before the next standup. The cost is concrete: you can win the conversion test and lose the money. A paywall that funnels more people into a free trial or a monthly plan can beat an annual-first one on Day-0 conversion and still earn less over the year, because some of those trials never convert and some of those monthly subscribers would have committed to a year if you hadn’t made the short option the easy one. You optimised the click, not the contract. This is exactly why we lean on predicted LTV in Adapty. You don’t have to wait a year to see which variant builds more long-term value; you get an early estimate in weeks. It’s still an estimate, but it moves the decision from “which paywall converts today” to “which paywall is worth more over time,” which is the question that actually matters.
4. With limited traffic, no team can test every idea at once, which is part of what you built Adapty’s Autopilot to solve. How does it decide what’s worth testing next?
Maxim: The core idea is that the test isn’t the product; the diagnosis is. Autopilot benchmarks a paywall against its category and comparable apps, then finds where that specific app underperforms its peers: trial start, trial-to-paid, plan mix, pricing, localisation. It ranks those gaps by expected revenue impact relative to effort and turns the top one into a concrete, ready-to-run A/B test rather than a vague recommendation. With limited traffic, you can’t test everything, so the only question that matters is which single test has the highest expected value right now, given where this app actually leaks. Geo-pricing is a big part of the answer there: one app across five countries is five parallel low-risk tests that never touch your main market, so even a low-traffic app gets several shots at once. It behaves more like a growth manager who prioritises than an engine that spits out variants.
5. Across 105,000+ paywalls, certain experimentation mistakes recur. What’s the most common way teams accidentally sabotage their own experiments: premature conclusions, bad traffic splits, testing too much at once?
Maxim: Premature conclusions, by a wide margin. Reading Day-0 or Day-3 numbers as the verdict when subscription outcomes take weeks to settle, and stopping on the first significant-looking result. Close behind is the opposite of over-testing: teams that never run the test that would actually move revenue because they’re scared of it, usually price. They’ll test a button colour for the tenth time and never touch pricing, so they spend their traffic on the safest, lowest-impact ideas. Third is running too many things at once on traffic that can’t support any of them, so nothing reaches significance and the team decides testing doesn’t work for them. The mechanical failure people worry about, bad traffic splits, is actually the rarest, because the platform handles that. The sabotage is almost always in the interpretation or the nerve, not the setup. The math is fine. The patience and the willingness to touch the scary lever isn’t.
6. AI is doing the heavy lifting in analysing funnels and surfacing patterns. As AI takes over more of the analysis, where does human product judgment still matter most in monetisation decisions?
Maxim: Two places. First, deciding what’s worth doing at all. AI can rank tests by expected impact, but choosing what you’re actually willing to try and what you won’t do to your users for a short-term win remains a human call. A recommendation can be numerically right and still wrong: adding a trial to a premium product, or cutting a price your customers read as a signal of quality. Takes a human five seconds to catch; an agent would just ship it. Second, the trust layer, especially when AI sits near money. When we put AI directly on top of pricing and revenue, people trusted it less, not more, because the stakes are real. So validation has to stay on the human, not as a formality, but because the paywall is the single most important screen in the app, the one where the revenue actually happens. That’s the last place you hand the keys to an agent and walk away. The way I put it in a recent talk: when you’re building AI near real money, don’t ask what the AI should do, ask what your users are afraid to do alone.
7. Working with experimentation data at this scale tends to reshape how someone thinks about risk and certainty. After seeing all these patterns, has your intuition about what makes a paywall work changed, and do you trust it more or less than you used to?
Maxim: Both, in opposite directions. On specifics – layout, copy, which plan goes first – I trust my intuition less than I used to. The data embarrasses everyone’s taste regularly, and design turns out to be one of the last levers, not the first. On the meta level, I trust it more. I’m now confident that structure and pricing beat aesthetics, that the bottleneck is almost always human rather than technical, and that a team that tests systematically beats a team with better instincts. So the direction of my confidence flipped: less faith in “I know what this paywall needs,” more faith in “I know what the process that finds the answer looks like.”
Why Trust Matters More Than Test Speed
Borisik’s view of subscription growth is clear: the hardest part of experimentation is not always infrastructure, traffic, or statistics. It is trust. Teams need to know which signals matter, which tests are worth running, and when short-term conversion wins may actually hurt long-term revenue.
As AI becomes more involved in paywall analysis and revenue optimisation, human judgment remains essential in deciding what should be tested, what should be avoided, and what kind of monetisation strategy users can still trust. Faster tools can raise the ceiling, but disciplined teams, better questions, and the courage to test meaningful changes are what turn experimentation into real growth.
Share
Copied
