A churn rate on its own tells you that people are leaving. It doesn't tell you who, when, or why. Without those three, every proposed fix is a guess.
At Travelxp I set up the metrics we used to track subscriber retention on the OTT platform. The most useful thing wasn't the headline churn number. It was splitting churn into pieces small enough to do something about.
Split voluntary from involuntary
The first split is the most important one.
Voluntary churn is when someone decides to cancel. Involuntary churn is when a payment fails and the subscription lapses, even though the person may have wanted to stay.
They need completely different fixes. Voluntary churn is a product and value problem. Involuntary churn is mostly a payments problem: expired cards, bank declines, failed renewals. Better retry logic and reminders to update payment details can recover a lot of it. I wrote more about that in the payment gateway post.
If you only look at total churn, a payments fix and a product fix get mixed together, and you can't tell which one worked.
Look at when people leave
| Timing | Common causes | Where to look |
|---|---|---|
| End of free trial | Didn't find enough value, forgot they signed up | Trial experience, first-week viewing |
| First paid month | Converted but didn't build a habit | Viewing frequency after conversion |
| After a specific event | Finished a series, a price change, a bad release | Content calendar, release history |
| Long-term subscribers | Gradual decline in use, or a better alternative | Usage trend over months |
Cohort analysis is what makes this visible. Group subscribers by when they joined and track how many remain over time. If newer cohorts drop off faster than older ones, something changed. The free trial post covers the earliest of these drop-off points.
Find the leading signals
Cancellation is the last step. Usually there are signs before it:
- Viewing frequency dropping over several weeks.
- Sessions that end without a play.
- Fewer platforms used, like stopping on TV and only using mobile.
- Repeated playback errors.
These signals let you act before someone cancels. That might mean better recommendations, fixing a playback problem on their device, or simply reminding them about something new they might like.
I'd look at the viewing pattern of people who cancelled in the weeks before they did. The patterns are often clearer than you'd expect. The quality metrics in the OTT streaming metrics post are a good place to start.
Use cancellation surveys carefully
Asking why someone is leaving is worth doing. Just treat the answers with some caution.
"Too expensive" is the most common answer in many services. Sometimes it really is about price. Often it means "not worth the price to me," which is a value problem. People also pick the quickest option to get through the survey.
I'd combine survey answers with behaviour data. If someone says "too expensive" but hadn't watched anything for a month, the real issue is probably engagement.
Prioritise by what you can change
Some churn is natural. People's circumstances change, and no product keeps everyone forever. The goal is to find the churn that's caused by something you can fix.
Rank the causes by how many people they affect and how practical the fix is. Fixing involuntary churn is often the quickest win. Improving discovery for subscribers whose viewing is slowing is usually a bigger, longer piece of work.
After any change, compare retention for affected cohorts against similar cohorts that weren't affected. Churn moves slowly, so give it enough time before deciding whether the change worked.