Cohort analysis groups your customers by the month they joined, then tracks each group separately over time. Instead of one blended average, you see how the January group behaves in month three compared with the June group. It separates the effect of your product changes from the effect of growth.
You build a cohort table with joining months as rows and months since joining as columns. Each cell shows how many of that group are still active. Reading down a column tells you whether newer customers behave better than older ones.
An example. Your January cohort has 200 customers. After three months, 120 still pay, so month-three retention is 60 percent. Your June cohort has 500 customers, and after three months 350 remain, which is 70 percent. Something you changed between January and June worked. A blended retention number across all customers would hide this, because the larger June group pulls the average around.
Cohorts also expose the opposite problem. Fast growth makes overall churn look low, simply because most customers are new and nobody has had time to leave yet. When growth slows, the churn that was always there becomes visible.
The mistake founders make is comparing cohorts at different ages. Your June cohort has existed for three months and your January cohort for eight. Of course more of the January group has left. Compare month three with month three, month six with month six, and never hold a young cohort's numbers against an old one's full life.
You do not need a large customer base for this. With 50 customers a month you can already see whether month-two retention moves. Cohort tables also work for revenue, order frequency, and usage, not only for retention.
