Late subscriptions can be the model, not a failure
Some subscription apps need revenue immediately. Others can build value over time and convert when a new need appears. Flo's growth lead Mike Charniatsou describes the second approach: keep the core cycle-tracking experience free, watch for changing needs, and introduce paid value when it becomes relevant.
A delayed conversion can protect the relationship
Flo reports that 40% of subscriptions occur more than six months after install. For many businesses that delay would signal weak early monetisation. At Flo, it reflects a deliberate long-term model: a user's needs and willingness to pay may rise, fall, or return as life circumstances change.
The free core product tracks cycles. Paid offers become relevant at later moments, such as trying to conceive, pregnancy, or perimenopause. Rather than pushing payment on day one or day seven, Flo waits until it can add something beyond the core experience. Charniatsou presents this as a luxury rather than universal advice. A younger app that must recover acquisition costs immediately may not have the same room to wait.
Judge the whole lifecycle, not just the paywall
Charniatsou's central warning is that paywall conversion alone does not define monetisation. Subscription performance unfolds across trial starts, trial activation, retention, and later behaviour. A change that lifts the first step can therefore weaken the business overall.
One example involved removing the close control from a paywall: conversion rose, but refunds rose too, turning the apparent win into a drain. His recommended discipline is to keep analysing shipped experiments and, at year end, compare the sum of reported uplifts with actual business growth. Any gap reveals how many apparent wins never became business wins.
Use this checklist before declaring an experiment successful:
- Did trial activation and retention move with initial conversion?
- Did refunds, cancellations, or complaints offset the gain?
- Does the cumulative reported uplift match real business growth?
- Was the result checked again after rollout, rather than only during the test?
Evidence beat explanation in a failed onboarding test
A perimenopause onboarding step initially looked logical but reduced conversion by almost 10%. The team repeated the test hoping the design was at fault and got the same result. Only then did it inspect the audience and discover that almost 75% of users could not be assigned a phase because the question was too difficult to answer.
Those users had arrived through an advertisement promising a phase result, yet received a dead-end message saying one could not be determined. Flo rewrote the question with its medical team and changed the logic so unplaced users continued to the next section. The evidence-based third version produced a 23% win. A later blackout test, which removed the new section, lost and helped confirm that the gain was real.
The practical distinction is between explanation and evidence. An explanation can justify almost any next iteration. Evidence shows where to investigate. Charniatsou's rule is to iterate when new evidence appears, not merely because a plausible explanation is available.
Marketing can surface needs before product does
One strong Flo onboarding branch began with advertising rather than a product roadmap. The acquisition team saw narratives about enhancing sex life gaining traction on Meta, with promising numbers even through the existing funnel. Product examined that signal, identified an unexplored user need, and built a dedicated branch that acknowledged the originating advertisement and asked tailored questions.
The broader lesson is organisational. Marketing can test audience messages at a scale and speed that a product funnel often cannot. Flo therefore shares marketing changes weekly across teams, even when their relevance is unclear, because insights that look minor to one team may matter greatly to another. Charniatsou does not describe the relationship as effortless; he describes it as necessary for a growth engine stronger than either team alone.
Borrow the problem, not the proven-looking solution
Charniatsou has repeatedly been offered other companies' successful paywall changes as ready-made solutions. Copying them pixel-for-pixel did not reproduce the results. His rule is to borrow the idea, not the solution: understand the underlying problem first, then design for your own audience.
A Blinkist-style paywall that reminds users before a trial ends is his clearest example. It has worked for other apps, but tests at Loóna and Flo did not work. A trial toggle produced the same outcome. The conclusion is not that those patterns are inherently poor; it is that another product's result is not evidence for yours.
The full episode also covers streaks as a possible cargo cult, proxy events for making early experiment calls when conversion is delayed, and placing partner mode into onboarding as Charniatsou's biggest fail of the year. Partner mode has more than 2 million pairs using it, which makes the onboarding failure a useful reminder that a popular feature is not automatically a good onboarding step.
Guest links
Mike Charniatsou on LinkedIn Flo Health website Flo Health Careers
Frequently asked questions
Why does Flo wait months to convert some users?
Flo reports that 40% of its subscriptions happen more than six months after install. It treats that timing as part of a long-term relationship, because needs and willingness to pay can change as users move through life stages.
How can a winning paywall test still be a business loss?
A paywall test can raise initial conversion while harming later trial activation, retention, or refunds. Check whether reported experiment gains add up to actual business growth over the year, and keep analysing results after rollout.
What changed in the perimenopause onboarding test?
After two losing runs, the team examined the audience and found almost 75% of users could not be assigned a perimenopause phase. Rewriting the question with medical input and routing unplaced users onward produced a 23% win, later checked with a blackout test.
Why not copy a paywall that worked elsewhere?
Borrow the underlying problem and reasoning, then design a solution for your own users. Copying a proven paywall pattern pixel-for-pixel failed at both Loóna and Flo in Mike Charniatsou's experience.