What is cohort analysis?
Cohort analysis is a way of studying behavior by grouping users who share a starting point or an action, such as signing up in the same month, and tracking each group over time. It shows whether newer users behave better or worse than older ones.
How do you do a cohort analysis?
You do a cohort analysis by splitting users into groups that share a starting point, choosing one behavior to track, and comparing each group's numbers at the same age. Comparing at the same age is the point: a user from January and a user from June are compared at their first month, not on the same calendar date.
The steps:
- Pick the cohort definition. Usually the week or month a user signed up or became a paying customer.
- Pick the metric. Most often retention (did they come back and do the core action?), but it can be revenue, feature use or upgrades.
- Pick the interval. Days, weeks or months, matched to how often the product should be used.
- Build the table. One row per cohort, one column per interval since start.
- Read across, then down. Across a row shows how one cohort ages. Down a column shows whether newer cohorts do better at the same age.
What are the types of cohorts?
Acquisition cohorts group users by when they started. They answer "are we getting better over time?" and catch the effect of onboarding, pricing or product changes.
Behavioral cohorts group users by what they did, such as "connected an integration in week one" or "invited a teammate". They answer "which early actions go with users who stay?" Comparing a behavioral cohort with everyone else is how teams find the actions worth pushing new users toward. That shows a link, not proof of cause, so test it before you build onboarding around it.
You can also cut acquisition cohorts by channel, plan or company size to see which segment is driving a change.
How do you read retention by cohort?
Here's a monthly retention table for a reporting tool, where "retained" means the user built or shared a report that month. Each cell is the percentage of the cohort still retained.
| Sign-up month | Users | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| January | 1,000 | 40% | 30% | 27% |
| February | 1,200 | 41% | 31% | 28% |
| March | 900 | 52% | 41% |
Read across January: a big drop after month one, then it levels off near 27%. Read down month one: January and February look the same, then March jumps. Something changed for the March cohort, say the team removed a setup step or prefilled a sample report.
Two checks before you celebrate. First, does the gain hold in month two (41% against 30% and 31%) and, once the data is in, month three, or did it only delay the drop? Second, was March smaller or different in mix (a new channel, a promotion)? Cut the cohort by segment to make sure the change, not the mix, moved the number.
What can cohort analysis not tell you?
Cohort analysis tells you which group changed and when. It doesn't tell you why. For that you need to see what users did and what they were trying to do: session recordings, interviews, and the questions users ask inside the product.
Foldspace covers both halves. Product analytics (funnels, retention and session replay) shows which cohort moved, and Conversational Analytics adds the why: the intent behind each request to the Product Agent, as users phrased it, and which conversations went unresolved.