Your monthly active users are growing. Your revenue is up. Everything looks healthy in the aggregate. But when you break the numbers down by when users joined, a different picture emerges. Users who signed up six months ago are highly active. Users who signed up last month are dropping off faster than any previous group. Your overall metrics are being propped up by early adopters while recent users are churning faster than ever.
Cohort analysis is the antidote to misleading averages. By grouping users based on when they started and tracking each group separately over time, you can see whether your product is genuinely improving or whether aggregate metrics are hiding a decline. It is the difference between looking at a snapshot and watching a movie.
The Core Idea
A cohort is simply a group of users who share a common characteristic, most commonly their signup date. A time-based cohort analysis tracks what happens to each group over the same intervals: Day 1 retention, Day 7 retention, Day 30 retention, and so on. By laying cohorts side by side, you can see whether newer groups perform better or worse than older ones.
You can also create behaviour-based cohorts — users who completed onboarding versus those who did not, users who activated a specific feature versus those who did not. These cohorts help you understand which behaviours predict long-term success and which segments need different treatment.