What is a cohort in analytics?

Why cohorts are necessary

Aggregate metrics blend users at different stages of their relationship with a product, so they move for reasons that have nothing to do with what changed.

The classic example: a product's overall retention rate drops. Nothing about the product got worse — a large marketing campaign brought in many new users, and new users always retain worse than established ones. The aggregate fell because the mix changed.

Cohort analysis eliminates that. Group users by when they arrived, follow each group forward separately, and compare like with like.

What cohorts reveal

• Whether the product is improving. Later cohorts retaining better than earlier ones is the clearest evidence available.
• Which acquisition sources produce value, since cohorting by source is the only way to see it.
• When users leave, which points at the specific stage that needs work.
• Whether a change worked, by comparing cohorts before and after it shipped.
• [LTV](/resources/glossary/lifetime-value-ltv) by group, which is what makes acquisition budgets defensible.

Reading them properly

Two disciplines separate useful cohort analysis from decorative charts.

Compare equal ages, not equal dates. A cohort's day 30 must be compared with another cohort's day 30, not with a calendar month. Recent cohorts have not lived long enough to have a day 30 at all, and treating an incomplete cohort as complete manufactures a decline.

Watch for confounders. A cohort acquired during a seasonal peak differs from one acquired in a trough for reasons unconnected to the product. Cohorting by source as well as by date usually separates them.