Cohort analysis in plain English

A cohort is a group of people who share something at the same starting point. In startups it is usually the month or week in which customers first signed up or first bought. The customers who joined in March form the March cohort, those who joined in April form the April cohort, and so on.

Cohort analysis is the habit of tracking each such group separately as time passes. Instead of asking how many customers are active today, it asks how many of the March cohort are still active one month later, two months later, six months later. Because each group is followed from its own starting date, the groups can be compared fairly.

A startup's overall numbers mix old and new customers. If the company is growing fast, most of its customers are new, and new customers behave differently from old ones. The overall retention figure can look fine even when each new group is leaving faster than the one before. It can also look weak just because a flood of new users arrived this month.

Cohorts remove that mixing. They let a founder see whether changes to the product, price or marketing made later groups better or worse. They let an investor see whether the business improves with age or only looks good because of a growth spurt.

A cohort table has one row for each starting month and one column for each month since joining. The first column is month zero, when everyone in the group is counted as 100 per cent. Each later column shows what share of the group is still active, or how much money the group has spent so far.

Read it in two directions. Reading across a row shows how one group ages: usually a sharp drop in the first month or two, then a flatter line. Reading down a column shows whether recent groups do better than older ones at the same age. If the month-three figure keeps rising from one row to the next, the product is getting stickier.

• Rows: the groups, by joining month.
• Columns: months since joining.
• Cells: share of the group still active, or money spent per customer.
• A line that flattens means a core of customers who stay.

The most common measure is retention: the share of a group that comes back or keeps paying. A second is revenue per customer over time, which shows whether customers spend more as they stay. This is the basis for estimating customer lifetime value, covered in the guide on CAC and LTV. A third is payback: how many months of a cohort's spending it takes to cover the cost of acquiring it.

For subscription software, a cohort can show whether customers upgrade or cancel. For a delivery app it can show how often each group orders in its fifth month. For a lending startup it can show what share of each month's borrowers repay on time, which is how credit quality is tracked.

Three shapes appear again and again. In the first, retention drops and keeps falling towards zero: the product is not sticky and the business keeps losing the customers it wins. In the second, the curve drops and then flattens at a level above zero: a loyal core stays, and the business can be built around it. In the third, the curve dips and then rises, usually because customers spend more or come back after trying alternatives. This last pattern is rare and valuable.

Also watch the first-month drop. Many consumer apps in India lose a large share of installs in the first days, so the useful question is what the survivors do afterwards.

Do not compare cohorts that are too small. A group of 40 customers can swing wildly with a few people. Do not mix different types of customers, such as those who came through a discount campaign and those who came through word of mouth, without splitting them. And define active clearly: opening the app, placing an order and paying a bill are all different things.

Finally, be careful with the newest rows. Recent cohorts have had less time, so they cannot yet be compared with old ones past the same age. Use the numbers to ask better questions, not as proof. This guide is educational and not investment advice.