What is cohort analysis?
Cohort analysis groups records by a shared starting event and then follows each group separately over time. A cohort is that group: everyone who signed up in March 2026, everyone who made a first purchase during a promotion, everyone hired in Q1. The defining feature is that membership is fixed at the start and never changes — a March cohort is always the same people, however many of them are still around.
That makes a cohort different from a segment, which is a slice defined by an attribute (enterprise plan, EU region) and whose membership can change as records change. You often combine the two: retention curves for the March cohort, split by plan.
Why cohorting beats an aggregate average
A single blended number mixes customers who have been with you for three years with customers who signed up yesterday, so it moves for two reasons at once: the underlying behavior changed, or the mix changed. Fast growth is the worst offender. If you double your new-signup volume, a company-wide retention rate will drop even if nothing about the product got worse, simply because young accounts are a larger share of the base. It also works in reverse — slowing acquisition makes retention look like it improved.
Cohorting removes the mix effect. Each cohort is measured against its own starting size at its own age, so month 3 for the January cohort is comparable to month 3 for the June cohort. That’s what lets you answer the question you actually care about: is the product getting better for new customers over time?
Reading the triangle
The standard display is a triangle (or heatmap): one row per cohort, one column per period since the cohort started, each cell showing the share still active. It’s triangular because recent cohorts haven’t lived long enough to fill the right-hand columns.
Read it two ways:
- Across a row — the retention curve for one cohort. Steep-then-flat means you have a durable core and an onboarding problem. A steady slide toward zero means no core at all.
- Down a column — the same age across cohorts. If month-3 retention climbs as you move down, newer cohorts are doing better, and whatever you changed is working.
Watch cohort size too. A tiny cohort’s percentages are noisy, and one big customer leaving can look like a collapse.
To build the triangle from your own data, start with the cohort retention dashboard.
Related terms
Put it to work
- Cohort retention dashboard — Dashboard
- Product retention dashboard — Dashboard
- Repeat purchase rate — Metric
