Cohort analysis

Cohort analysis groups customers by a shared starting event, most commonly the month or week they signed up or made a first purchase, and then follows each group forward in time. Instead of looking at total active users in a single month, it asks how the January cohort behaved in its second, third and twelfth month.

Analysts typically arrange cohorts as a triangular table, with each row a cohort and each column a period since the starting event, showing a metric such as retained users, revenue, or churn rate at each point. This makes it possible to see whether a product or onboarding change made later cohorts perform better or worse than earlier ones, something a single aggregate trend line would hide.

Cohort analysis matters because it separates genuine changes in customer behavior from shifts in the mix of who is joining, and it is the standard technique behind retention analysis and predictive customer lifetime value models. The main pitfalls are drawing conclusions from cohorts too small to be reliable, and survivorship bias, where the customers still visible in later periods are, by definition, the ones who did not leave.