Cohort Analysis Template for Product Analytics

The retention event should measure whether the user got value from the product, not just whether they showed up. For a project management tool, "completed a task" is better than "logged in." For a messaging app, "sent a message" is better than "opened the app." The test: if a user performed only this action and nothing else, would they have gotten value? If yes, it is a good retention event. See the [glossary entry on cohort analysis](/glossary/cohort-analysis) for more on choosing the right event.

Weekly cohorts for products with daily or weekly usage patterns (SaaS tools, productivity apps, social products). Monthly cohorts for products with lower-frequency usage (marketplaces, financial tools, seasonal products). The rule of thumb: your cohort window should be shorter than your natural usage cycle. If users typically return weekly, monthly cohorts will miss important trends.

At minimum, 6-8 cohorts of similar size. With fewer, random variation dominates and trends are unreliable. For behavioral segments, each segment needs at least 100 users per cohort to be statistically meaningful. If segment sizes are too small, widen your cohort window (e.g., monthly instead of weekly).

Pick a clean cutoff date (the date the change shipped to 100% of users). Cohorts before that date are the "before" group; cohorts after are the "after" group. Compare the same retention windows (D7, D30) across groups. Allow at least 3-4 post-change cohorts before concluding the change worked, to account for novelty effects and seasonality.

Overall retention blends all users regardless of when they signed up. It is slow-moving and hides trends. Cohort retention isolates each signup group, so you can see whether your product is getting better at retaining new users over time. A product can have declining overall retention (because of a growing base of churned users) while having improving cohort retention (because each new cohort retains better than the last). Cohort retention is the metric that tells you whether your improvements are working. ---