Cohort analysis is a method of grouping users by a shared characteristic or starting point in time — most commonly their acquisition date — and then tracking that group's behavior, such as retention or spending, over subsequent periods.
The core technique is comparing cohorts against each other rather than looking at a blended, aggregate number: a table might show what percentage of the 'January signups' cohort was still active in month two, three, and four, laid alongside the same figures for 'February signups'. This makes it possible to see whether a product change improved retention going forward, something a single averaged retention figure would obscure.
A subscription app might build a cohort table showing that 60% of users who signed up in March were still active after 30 days, compared to 45% of users who signed up in January — suggesting something changed for the better between those two months.
By acquisition date — the week or month a user first signed up — though cohorts can also be built around shared behavior, like a first purchase, or shared attributes like acquisition channel.
Retention rate is typically a single aggregate figure; cohort analysis breaks that figure apart by group so you can compare how retention differs across different starting points or segments over time.
Because it lets you compare a cohort that experienced a change to one that didn't, isolating whether behavior actually improved for new users rather than looking at a blended number that mixes old and new users together.