Cohort Analysis
Cohort analysis is the practice of grouping users by a shared characteristic (usually signup date) and tracking how each group behaves over time.
What it means
When you look at total active users, you see one number changing over time. That hides the real story. Cohort analysis splits users into groups (cohorts) based on when they signed up, then tracks each group separately. You can see whether new cohorts retain better than old ones, which is how you know if your product is improving.
A typical cohort analysis is a grid: rows are signup months, columns are weeks since signup, cells show what percentage of that cohort was active. The first cell is always 100%. Each cell to the right shows how many stuck around.
If you ship a major change to onboarding in March, the cohorts after March should retain better than cohorts before March. If they don't, your change didn't work. Cohort analysis is the only way to see this clearly because aggregate numbers smooth over the differences.
Why it matters
Cohort analysis is the highest-resolution view of whether your product is getting better. Aggregate metrics lie because they mix old and new users. Cohort metrics show whether each new generation of users does better than the last. Without cohorts, you're flying blind on retention.
Example with real numbers
Concrete example showing how this metric works in practice.
Scenario
You shipped a new onboarding flow in March. You compare your March cohort retention to your January cohort retention.
What it means
January cohort: 30% week-4 retention. March cohort: 45% week-4 retention. Your onboarding change worked. New users are sticking around 50% better than before.
Common mistakes
Things people get wrong when measuring cohort analysis.
Mistake 01
Cohort sizes too small to be meaningful. With 20 users in a cohort, the noise drowns out the signal.
Mistake 02
Comparing cohorts from very different periods. Seasonal effects can make April cohorts look worse than November cohorts even if nothing changed.
Mistake 03
Looking at cohort tables without acting on them. The point is to compare cohorts before and after specific changes.
Mistake 04
Only making cohorts based on signup date. You can also cohort by acquisition channel, plan, or other meaningful traits.
How to track it
Group your users by signup week or month. For each cohort, track what percentage was active in each subsequent week. Most analytics tools, including Muro, can build cohort tables automatically once you define an active event.
Free tools to help
Muro built free calculators and analyzers around this metric.
Related concepts
Other terms worth learning if you're studying this one.
Common questions about cohort analysis
Cohort analysis groups users by when they signed up (or another shared trait) and tracks how each group behaves over time. It's how you see whether new users are sticking around better than older ones.
Total active users mix everyone together. If you ship a great improvement that makes new users 2x as sticky, total active users barely move (because most active users are old cohorts). Cohort analysis catches the improvement immediately.
Group users by signup month. Track what percentage of each group is still active at week 1, week 4, week 8, etc. Compare cohorts to see if newer ones retain better. The pattern reveals whether your product is improving.