What is Retention cohort?
A retention cohort is a group of users defined by when they first arrived, measured for how many return in each later period. Because every cohort is followed on the same clock — week one, week two, week three — it separates changes in product quality from changes in acquisition volume, which a single retention figure cannot do.
What the grid shows
Rows are cohorts by start period, columns are periods since. Reading across a row follows one group as it ages; reading down a column compares different groups at the same age.
Cohort Week 0 Week 1 Week 2 Week 4
Jan 06 100% 42% 31% 24%
Jan 13 100% 45% 33% 26%
Jan 20 100% 51% 39% —The column comparison is the one that answers “did the change work”. Week-1 retention rising from 42% to 51% across cohorts is evidence that something improved for people arriving later — and it is completely invisible in a blended retention number, which mixes cohorts of different ages together.
The shape of the curve matters more than its level
Retention curves fall steeply and then either flatten or continue to zero. The flattening is the whole question.
- A curve that flattens has found a group of people for whom the product is habitual. That plateau, not the starting height, is what compounds into a business.
- A curve that keeps falling means everyone eventually leaves, and growth is a treadmill: acquisition has to replace the entire base repeatedly.
Two products can have identical week-1 retention and opposite futures.
Practical notes
- Cohorts need volume. Weekly cohorts of 30 people are noise. Widen to monthly until each cohort is large enough to see through.
- “Returned” needs a definition. Opened the site, performed a meaningful action, and paid are three different curves. State which one.
- Segment by acquisition channel. Retention varies enormously by where people came from, and a blended curve hides the channel sending users who never come back — which is exactly the channel you should stop paying for.