How does cohort analysis work?
You group users by when they first showed up — the January signups, the February signups — then measure the same behavior for each group at the same relative age: day 1, day 7, day 30. The result is usually a grid where each row is a cohort and each column is an age. Reading down a column answers the question a blended average cannot: do newer cohorts retain better than older ones, or is the product standing still while acquisition grows?
How do you read a retention curve?
Plot one cohort's retention rate against time since signup and you get a curve that starts at 100% and falls. The shape matters more than any single point. A curve that keeps sliding toward zero means every user eventually leaves — growth is refilling a leaking bucket, and your churn rate will catch up with acquisition. A curve that flattens means a core of users has settled in. Growth practitioners commonly read a flattening curve as a product-market-fit signal; it is a heuristic rather than a formal test, but a widely trusted one.
Why it matters for a two-person team
Blended metrics flatter growing products: when this month's signups outnumber last year's, the average is dominated by users too new to have left. Cohorts strip that distortion out with one grouping — and the query is simple enough to run in SQL against your own events. Pair the day-30 column with your activation rate to check whether onboarding changes actually stick. What to do when a cohort sags — win-back journeys, lifecycle messaging — is covered in the guide to retention and win-back flows.
What are the common mistakes in cohort analysis?
Three recur. Cohorts too small to be signal — in a 20-user cohort, one person moves retention by five points, so read the trend across several cohorts rather than any single one. Comparing cohorts at different ages — a two-month-old cohort always looks better than a twelve-month-old one at month twelve, because it has no data there yet. And defining 'active' loosely — if opening an email counts as retention, the curve flatters you. Tie 'active' to the action that delivers your product's core value, not to a login.