What Is Cohort Analysis? A Plain-English Guide

Short answer: cohort analysis groups people who share a starting event in the same period, usually the week or month they signed up, and follows each group over time. It shows whether newer users behave better or worse than older ones, which a single overall average cannot.

Table of Contents

1. Why Averages Hide the Truth
2. How to Read a Cohort Table
3. Types of Cohorts
4. What a Healthy Curve Looks Like
5. How to Build One Simply
6. From Cohorts to Causes
7. Key Takeaways

Why Averages Hide the Truth

Suppose your active users are growing. That could mean people are staying, or it could mean you are adding new users faster than old ones leave. The total looks identical. Only by following each signup group separately can you tell.

The same applies in reverse. Improve onboarding in June and the overall retention number barely moves, because it is dominated by everyone who signed up earlier. The June cohort will show the change immediately.

How to Read a Cohort Table

An illustrative retention table:

Read it three ways.

• Across a row: how one group decays over time. January fell from 40% to 24%.
• Down a column: whether newer groups do better at the same age. Month 1 retention rose from 40% to 53%. Something improved in March.
• Along the diagonal: what happened in one calendar month across all groups. A dip along a diagonal points to an outage or a seasonal effect.

Types of Cohorts

Behavioral cohorts are the most useful for product decisions. "Users who invited a teammate in week one" against "users who did not" tells you which early actions matter.

What a Healthy Curve Looks Like

Plot one cohort's retention over time.

• Declining to zero: people try the product and all eventually leave. No amount of acquisition fixes this.
• Flattening: the curve drops and then levels off above zero. A core group keeps finding value.
• Rising after a dip: some users return, which is rare and a strong sign.

A flattening curve is one of the clearest behavioral signals of product-market fit (more on this).

How to Build One Simply

You do not need a specialised tool to start.

1. Choose the starting event. Usually signup.
2. Choose the return event. Not "logged in" but your core action: sent a message, created a report, placed an order.
3. Choose the period. Weekly for products used often, monthly for products used occasionally.
4. Export two columns: user ID with signup date, and user ID with activity dates.
5. Build a pivot table of signup period against periods since signup.

From Cohorts to Causes

A cohort table tells you that March users retained better than February users. It does not say why. That is where individual sessions help.

1. Pick the cohort that changed.
2. Identify users in your recordings so you can find people by ID.
3. Watch first sessions from retained users and from churned users in that cohort.
4. Look for what the retained group did that the others did not.

With DeepSync, calling identify after login links every past and future session to your own user ID, so a cohort from your database can be looked up as real sessions (SDK API):

deepsync("identify", user.id, { signupDate: user.createdAt, plan: "trial" });

DeepSync does not replace a dedicated cohort report. It supplies the evidence for why a cohort behaved as it did.

Key Takeaways

• Cohorts follow groups with a shared start over time.
• Read rows for decay, columns for improvement, diagonals for events.
• A curve that flattens above zero is the healthy shape.
• Use recordings of retained and churned users to learn why.

Conclusion

An overall retention number blends every user you have ever had. Cohorts separate them, and that is where you find out whether the product is getting better.

Start free with DeepSync.