second open
← All case studiesPrevious work · 2024 to 2025

AI learning app at 30M users

A European AI study companion used by students in 20 countries. As product engineers inside the team, we shipped more than ten features, helped open a new market, and built monetization features that helped cut churn by 35%.

Role
Product engineer, inside the product team
Surface
iOS and Android
Reach
30M+ students, 20 countries
Company
Unnamed here by choice
−35%churn, helped by the monetization features we built
30M+students using the product
20countries, and one new market opened
10+features shipped to students worldwide
The challenge

Growth was never the issue. Keeping students past the first month was.

A study app has a rhythm: exams pull students in, holidays push them out. On a subscription product that rhythm shows up as churn, and at 30 million users one point of churn is a lot of people. The monetization work had to keep paying students through the quiet weeks without making the free product worse.

What we did

Work on the moments where students decide to stay.

We worked on the subscription path end to end: where the upgrade is offered inside studying rather than in front of it, what a student sees as a trial ends, and what happens when they try to leave. Each change shipped on both platforms and was judged on renewals, not on the day's conversion.

Where our work sat

  1. 01
    The offer, inside the study flow
    Shown once a student has used what it unlocks, never as a wall in front of the feature.
  2. 02
    A trial that ends on a clear choice
    What the student keeps, what they lose, and a plain way to decide.
  3. 03
    The renewal, measured by cohort
    Every change judged on who renews, by signup week, not on the day's upgrade rate.
  4. 04
    A new market: Romania
    Language, content and store setup for the launch.

The four tests, on this project

We judge every product on the same four questions. This is how this one answered each of them.

01 · Design
Does it feel right?
  • Upgrade moments placed after a student has seen the value, never in front of it.
  • Offers built from the product's own components, so they read as the app and not an ad.
02 · Mobile and backend
Does it hold up?
  • More than ten features shipped on iOS and Android to 30 million students.
  • A new market opened: Romania, with its language, content and store setup.
03 · AI, retrieval and evals
Is it right?
  • Outside our role here. The tutor's models were another team's work; ours was the product and the subscription around them.
04 · Metering and pricing
Led here
Does it pay?
  • Monetization features that helped cut churn by 35%.
  • Judged on renewals and churn, not on the day's upgrade rate.

Three decisions we made

01
Churn by cohort, not by month
A monthly churn number mixes exam season with summer. Cohorts by signup week showed which changes actually kept students.
02
Fix the renewal before the funnel
More trials into a leaky subscription is a bigger leak. The work started where students cancel, then moved upstream.
03
Offer after value
An upgrade shown once a student has used what it unlocks converts, and renews, better than a wall in front of the feature.
What this means for you
If your AI feature gets tried once and dropped
The fix is rarely the model. It is usually where the offer sits, what the trial ends on, and what leaving looks like. Those are the first three things we check in week one, with your own cohorts, before anyone touches a prompt.
Closest track: Stop the churnSee the two weeks
What we can build for you
  • Subscription, trial and paywall flows on iOS and Android
  • Churn measured by cohort, not by month
  • Experiments judged on renewals, not first conversion
  • Launching an app into a new market
Stack
iOSAndroidIn-app subscriptionsExperimentsLocalisation

Want this in your product?

Thirty minutes, no deck. Tell us what you are building and which of the four tests worries you most. We say no when we are not the right people.

Book a 30-minute callNext: Multi-agent AI platform