Cohort Analysis

Cohort analysis is a method of grouping users who share a common characteristic (usually signup date, but also acquisition channel, plan type, or first action taken) and then tracking how each group behaves over time. Instead of looking at all users as one blob, you compare the January cohort to the February cohort to the March cohort to see if things are getting better or worse.

Why it matters: aggregate metrics lie. If your overall retention rate is 40%, that could mean every cohort retains at 40%, or it could mean your older cohorts retain at 60% while your recent cohorts retain at 15%, and the average just happens to land at 40%. Those two scenarios demand completely different responses. Cohort analysis reveals the truth beneath the average.

How to do it: the most common format is a retention cohort table. Rows represent cohorts (grouped by signup week or month). Columns represent time periods after signup (Week 0, Week 1, Week 2, etc.). Each cell shows the percentage of that cohort still active in that period. This creates a triangular matrix that instantly shows whether your retention curves are improving over time.

Tools for cohort analysis: Amplitude and Mixpanel have built-in cohort analysis features that let you define cohorts by any event or property. Google Analytics 4 offers a basic cohort exploration. For more custom analysis, many teams export data to a warehouse (BigQuery, Snowflake) and build cohort tables in a BI tool like Looker, Metabase, or even Google Sheets.

Common mistakes: choosing cohort windows that are too small or too large for your business cycle. For a daily-use app, weekly cohorts make sense. For enterprise software with quarterly contracts, monthly or quarterly cohorts are more appropriate. Another mistake is only looking at time-based cohorts when behavioral cohorts (grouped by first action taken, feature adopted, or channel) often yield more actionable insights.

Practical example: a mobile app groups users by signup month and discovers that March and April cohorts retain 25% better than January and February cohorts. The team traces this to an onboarding redesign shipped in late February. They now have quantitative proof that the redesign worked, and they double down on the new onboarding approach.