Cohort Analysis

Cohort analysis groups users by something they share — usually the month they signed up — and tracks how that specific group behaves week-over-week or month-over-month. The point is to separate the behavior of users who signed up in March from those who signed up in April, so you can see whether your product is actually getting better at retaining users or just hiding declining retention behind growth.

Why aggregate metrics lie

Look at “what percent of all users are active this week.” That number conflates users who signed up yesterday (very likely active) with users who signed up a year ago (likely churned). If you’re growing fast, the new users mask the old users churning out. Active-user rate looks fine right up until growth slows and the cohort decay catches up.

The cohort view shows you the underlying truth: of users who signed up in March 2024, what percent are still active 3, 6, 12 months later? Same question for the April cohort. If the curves are converging on a floor — call it 40% — that’s your true retention. If each newer cohort decays faster than the last, the product is getting worse for retention even if growth is masking it.

The standard cohort table

Rows are cohorts (Jan, Feb, Mar…). Columns are months since signup (M0, M1, M2…). Each cell is the percent of that cohort still active that many months in.

The diagonal shows you what month you’re in (only the most recent cohort has data for M0; only older cohorts have data for M12). The flatness or steepness of each row tells you whether retention is plateauing or still bleeding.

How Datost handles cohort analysis

Ask Datost “show me retention by signup cohort for users who signed up in 2025” and it returns a cohort table from your warehouse with the SQL attached. The hard part is usually figuring out what counts as “active” — Datost reads your metric definitions to use the same active-user definition your team has agreed on, so the cohort table you get matches the cohort table on your dashboard.