How do you predict churn in a subscription app?
Predict churn from behaviour: a user's activity falling against their own recent pattern, a core action they stopped doing, and subscription events such as auto-renew turned off or a failed payment. Score every user on those signals daily, and use the score to decide who gets a message and who is left alone.
Signals that predict churn
| Signal | Example | Why it works |
|---|---|---|
| Activity falling against the user’s own pattern | 10 sessions a week, now 2 | Drift shows up before the cancel |
| A core action stopped | Workouts logged fell to zero | Value has stopped, even if the app still opens |
| A broken streak or unfinished goal | A course two lessons from done, untouched for a week | The user is close to a result and stalled |
| Auto-renew turned off | Store event from your billing source | The user decided; access usually continues to the end of the period |
| Payment failed | Card declined at renewal | Involuntary churn, unless the user updates the card |
| Trial ending without the core action | Trial day 5 of 7, no core action yet | Nothing to pay for yet |
1. Track the right events
Prediction is only as good as the events behind it. Track the actions that show a user is getting value, with timestamps and a stable user id. Add subscription events from your billing tool. Skip events that fire on every screen; they add noise.
2. Compare users with themselves
A user who opens your app twice a week isn’t at risk if they always did. A user who opened it daily and now twice a week is. Score drift against each user’s own recent history: this week against the last four, for example.
3. Weigh the events that matter for your app
A meditation app and a budgeting app lose users in different ways. Give more weight to the actions that separate users who stayed from users who left in your own history. Review the weights when your product changes.
4. Score every user, every day
Recalculate the score daily so a change in behaviour shows up quickly. Keep the score per user, not per segment. Segments hide the user who was your most active last month and is quiet this week.
5. Turn the score into a decision
A prediction is useful only when it changes what you do. For each user the choice is: send nothing, send a message, or wait. Most users should get nothing on most days. Add caps, cooldowns and quiet hours so the score can’t flood anyone.
6. Check it against a holdout
Keep a random group out of all messages and compare their retention with everyone else’s. That’s the only way to know whether acting on your predictions keeps users. How to run a holdout test.
Where Whisperr fits
Whisperr scores how likely each user is to come back, from their own recent activity. The decision to send is a separate step: Whisperr weighs the product events that matter for your app, adds subscription events from RevenueCat, Superwall or Adapty (trial ending, auto-renew off, payment failed), and checks caps, cooldowns and quiet hours. Only when a message would help does it write one for that user and send it by push or email through your own providers. A holdout group shows how many users the messages kept.
FAQ
Do I need machine learning to predict churn?
Not to start. A simple rule that compares each user's activity this week with their own average in recent weeks catches most drift. Machine learning helps when you have many users and many signals, and it still needs the same clean events underneath.
What data do I need to predict churn?
Timestamped events for the actions that show value, a user id that stays the same across devices, and subscription events (trial, renewal, auto-renew off, payment failed). Demographics and acquisition source add little once you have behaviour.
How early can you predict churn?
Activity often starts falling weeks before a user turns off auto-renew, so behaviour gives you an early signal. How early depends on how often people use your app; a daily habit app shows drift sooner than an app people open once a month.
What should I do with a churn prediction?
Use it to decide who gets a message, what it says and when. A score that only fills a dashboard changes nothing. Keep a holdout group so you can see whether acting on the prediction actually keeps users.
How does Whisperr predict churn?
Whisperr scores how likely each user is to come back, from their own recent activity. To decide whether to message someone, it weighs the product events that matter for your app and the subscription events from RevenueCat, Superwall or Adapty, then applies caps, cooldowns and quiet hours. Many users get nothing.
Let Whisperr run this for you.
Whisperr watches how people use your app, spots the ones about to leave and brings them back with the right message at the right moment. After you sign up, one prompt in your coding agent sets it up.
You approve until you trust it. Then it runs itself.