Cohort Analysis
Quick definition
Cohort analysis groups users by a shared characteristic or starting point, such as the week they signed up or the campaign that brought them in, and compares how each group behaves over time.
Why use cohorts
Averages across all users can hide what is changing. A product may look stable overall while newer users behave differently from older ones. Comparing cohorts shows whether a change, such as a new onboarding flow, a price change or a different acquisition channel, affects the people who experienced it.
Common ways to define a cohort
- Acquisition date, such as sign-up week or month.
- Acquisition source or campaign.
- First product, plan or purchase.
- A behavior, such as completing onboarding.
Reading a cohort table
A cohort table puts cohorts in rows and time since they started in columns, so each cell shows what share of the cohort did something, such as returned or purchased, in that period. Reading across a row shows how one cohort changes with time. Reading down a column compares cohorts at the same age.
Limits
- Small cohorts produce noisy percentages.
- Define the cohort and the measured action before looking at the results.
- Differences between cohorts can come from outside factors such as season or promotions.
Example
A subscription app groups users by sign-up month. Of the 1,000 users who joined in January, 400 are still active in month 2 (40%). Of the 1,200 who joined in February, 600 are active in month 2 (50%). Something changed between the two cohorts and is worth investigating.