Cohort Analysis

Cohort analysis groups customers by when they started and watches each group age, instead of blending everyone into one number that hides the truth. A January cohort and a June cohort are tracked separately, month over month, so you can see whether the customers you signed last quarter behave better or worse than the ones you signed last year. Blended metrics tell you where you are. Cohorts tell you which direction you're going — and they're the only way to catch a product getting better or a sales motion getting worse before the aggregate number admits it.

How Cohort Analysis Is Calculated

Pick a cohort key (usually signup month), a metric (retention, revenue, expansion), and a time axis measured from each cohort's own start — Month 0, Month 1, Month 2. The output is a triangular table: each row is a cohort, each column is months-since-start.

Retention for a cohort at month N:

Cohort retention (Month N) = (Customers from cohort still active in Month N) ÷ (Customers in cohort at Month 0)

For revenue, swap counts for dollars and you get a revenue retention curve per cohort — the cohort-level cousin of net revenue retention.

A Worked Cohort Analysis Example

Three cohorts, one ugly story. Each new cohort retains worse at the same age — 88% vs 85% vs 79% at Month 3. Blended retention might still look fine because the healthy January cohort props up the average. The cohort view says the recent customers are leaving faster, which usually means sales started closing the wrong accounts. Same data, opposite conclusion.

When Sales Teams Use Cohort Analysis

RevOps and Finance own cohort analysis because it underwrites the LTV assumptions in every board deck — a flattening retention curve quietly destroys lifetime value math. A CRO uses cohorts to test whether a new sales playbook actually produced stickier customers or just more of them. Customer Success uses cohort curves to spot the month where churn spikes, then builds the intervention one month earlier. Investors ask for cohort tables specifically because they're the hardest metric to fake in a data room.

Common Cohort Analysis Misconceptions

The biggest error is reading early cohorts as destiny. A 12-month-old cohort looks great partly because the customers who were going to leave already did — survivorship makes old cohorts flatter than young ones ever will be. Comparing a mature cohort's Month 12 to a young cohort's Month 3 isn't a comparison; it's two different questions.

Cohort tables also get cropped. Show only the cohorts that retained well, hide the recent ones still bleeding, and the slide tells a story the churn rate would contradict. Always check that the newest cohorts are present — they're the ones that reveal whether today's selling is working.

And cohorts measure the customers you kept, not the ones you should have signed. A cohort can retain beautifully and still be the wrong cohort — small, cheap accounts that never expand. Pair retention curves with revenue cohorts and gross revenue retention, or a flat logo curve will flatter a shrinking book of business.