Cohort Analysis

Cohort analysis groups users by a shared starting point and tracks each group over time, comparing behaviour by age in the product rather than by calendar date.

The output is usually a triangular table: each row is a cohort, each column a period since the shared origin, and each cell the share of that cohort still performing the qualifying action. It reads in two directions. Moving down a column compares cohorts at the same age and shows whether the product is getting better or worse, while moving across a row shows the decay shape of a single cohort. The main value of the method is that it removes the mix effect: an aggregate metric can rise purely because new users keep arriving while every individual cohort is decaying. Two things break the reading. The newest cohorts have not lived long enough, so the right side of the triangle thins out, and small cohorts produce noisy percentages, which is why cohort size belongs next to every row.