Cohort Analysis: The Metric Type That Changed How SaaS Companies Think About Churn

If you run a SaaS company and you are looking at a single churn number — say, "our monthly churn is 4%" — you are flying blind. That number hides more than it reveals. Cohort analysis is how you see the real picture.

What Is a Cohort?

A cohort is simply a group of customers who share a common starting point. Usually, it is the month they signed up. The January 2026 cohort is everyone who became a customer in January 2026.

Cohort analysis tracks what happens to each group over time. Instead of asking "what is our churn rate?", you ask "of the customers who signed up in January, what percentage are still active in month 1, month 2, month 3?"

How to Read a Retention Table

A cohort retention table is a grid. Rows are cohorts (sign-up months). Columns are months since sign-up. Each cell shows the percentage of that cohort still active.

Example:

| Cohort | Month 0 | Month 1 | Month 2 | Month 3 | Month 6 | |----------|---------|---------|---------|---------|---------| | Jan 2026 | 100% | 82% | 71% | 65% | 52% | | Feb 2026 | 100% | 85% | 76% | 70% | — | | Mar 2026 | 100% | 88% | 80% | — | — |

Read it left to right: each row tells you the lifecycle of one group. Read it top to bottom: each column tells you whether things are improving over time.

The Three Patterns to Look For

1. The Flattening Curve

If a cohort drops from 100% to 70% in month 2 but then stays near 65% through month 12, you have a product that retains its core users well. Your problem is in onboarding and early activation, not in the core product. Fix the first 30 days.

2. The Steady Decline

If every cohort keeps dropping — 80%, 65%, 50%, 35%, 20% — with no flattening, you have a product problem. Customers are not finding lasting value. No amount of onboarding optimization will fix this.

3. Improving Cohorts Over Time

If the March cohort retains better than the January cohort at the same point in their lifecycle, something you changed is working. This is how you measure the impact of product changes, onboarding improvements, or pricing adjustments.

Why a Single Churn Number Lies

Imagine you had terrible churn in Q1 but shipped a major product improvement in Q2. Your overall churn rate blends old, poorly-retained cohorts with new, better-retained ones. The blended number makes things look worse than they currently are — or in the reverse case, better than they are.

Cohort analysis separates signal from noise. It tells you whether things are getting better or worse right now, regardless of what happened six months ago.

How to Build This

You need three data points per customer: sign-up date, activity dates (or payment dates), and cancellation date (if applicable).

The SQL is straightforward. Group customers by sign-up month, then for each subsequent month, count how many are still active. Divide by the original cohort size.

If you are in a BI tool like Looker, Metabase, or Tableau, most have built-in cohort visualization. In a spreadsheet, a pivot table gets you there, but it becomes painful to maintain past 500 customers.

What to Do With the Insights

Once you can see your cohorts clearly:

• **If early drop-off is steep:** Invest in onboarding. Add a check-in at day 7, day 14, day 30. Identify what "activated" users do that churned users do not.
• **If decline is steady:** Interview churned customers. The product is not delivering sustained value.
• **If recent cohorts are better:** Document what changed. Double down on it.

Cohort analysis is not a nice-to-have. For any subscription business, it is the single most important analytical tool you have. If you are not looking at cohorts today, start this week.