Users who started at the same time — typically installed or signed up on the same day or week — form a cohort. You group them this way to watch how that same set behaves as time passes.
Why you need it
Total active users can look like they're growing just because new users keep arriving, even while every cohort is quietly churning. Splitting users into cohorts and tracking each one over time reveals whether the product is actually holding on to people.
Where it's used
Channel evaluation: does a cheap channel also bring good retention?
LTV estimation: cumulative revenue curves per cohort are the basis of LTV modeling.
Go deeper
How to read cohort retention curves in practice is covered in Cohort Analysis.
Frequently asked questions
How is a cohort different from retention?
A cohort is the unit — users who started at the same time. Retention is the share of that group still returning later. Average retention that is not split by cohort moves with changes in new-user volume, which makes it hard to interpret.
What should cohorts be grouped by?
Usually install or signup date, though first purchase date is also used. Any choice works as long as the starting point is stated on the chart, and changing it midway breaks comparison with past periods.
Why do recent cohorts suddenly look worse?
Most often because the observation window has not filled. A user who installed yesterday has no D30, and rendering that as zero looks like a collapse. Leave immature cells blank or mark them separately.