Cohort analysis

What is cohort analysis?

Cohort analysis is a method that splits your users or customers into groups, called cohorts, based on a shared characteristic, then compares how each group behaves over time. The most common characteristic is the date of first action: everyone who signed up in January forms one cohort, everyone who signed up in February forms another. Looking at cohorts side by side shows whether your product, onboarding or marketing is getting better or worse.

It matters because averages hide change. A blended retention rate of 40% can mask the fact that recent users leave twice as fast as older ones. Cohorts make that visible.

‍

How a cohort table works

The standard output is a table. Rows are cohorts, columns are time periods since the cohort started, and each cell shows a metric such as the share of users still active.

Read the table in two directions. Across a row, you see how one cohort decays. Down a column, you compare cohorts at the same age, which shows whether a change you shipped improved retention.

‍

Types of cohorts

• Acquisition cohorts: grouped by sign-up or first purchase date.
• Behavioural cohorts: grouped by an action, such as using a feature in the first week.
• Channel cohorts: grouped by source, for example organic search versus paid.
• Segment cohorts: grouped by plan, country or company size.

‍

Cohort analysis vs segmentation

The two are related but answer different questions.

‍

How to run a cohort analysis

1. Define the question, for example: do users who complete onboarding stay longer?
2. Choose the cohort definition and the time unit (day, week, month).
3. Pick one metric: retention, revenue per user or conversion.
4. Build the table in your analytics tool or a spreadsheet.
5. Compare cohorts, find the break points and form a hypothesis.
6. Change one thing, then watch the next cohort.

A simple SQL starting point groups users by sign-up month:

SELECT DATE_TRUNC('month', signup_at) AS cohort, COUNT(*) AS users FROM users GROUP BY 1 ORDER BY 1;

‍

Pitfalls to avoid

Cohorts that are too small produce noisy results, so wait for enough volume. Mixing different definitions of active makes cohorts incomparable. Seasonality can make a December cohort look better or worse for reasons unrelated to your changes. Finally, a retention curve describes what happened but not why, so pair it with interviews or session recordings.

‍

Cohort analysis at BeBranded

We use cohort analysis in our consulting work to show where a product or website loses users, which acquisition channels bring lasting customers and which changes truly move retention. Discover our consulting service.