How cohort analysis works
A cohort is a group of customers who share a starting point — usually the month they first bought, though it can be the campaign that brought them in or the first product they chose. Instead of pooling everyone into a single average, you line the cohorts up beside each other and watch each one age.
The table that results has cohorts down the left and elapsed time across the top. Every cell answers the same question for a different group: of the people who arrived in this period, how many were still buying a month later, a quarter later, a year later, and how much had they spent by then. Because each cohort is measured from its own first day, a group that arrived in a quiet season is still comparable with one that arrived in a rush.
Why cohort analysis matters
Blended averages hide the thing you most need to know: whether the customers you are buying now are worth as much as the ones you bought before. Total revenue can climb while every new intake is quietly weaker than the last, because volume masks the decay. Cohorts show that while the acquisition budget can still be redirected.
It also tells you when your money comes back. If a cohort takes several months to repay what it cost to acquire, that is a cash-flow fact rather than an opinion — and for a business running on its own working capital, it decides how fast you can afford to grow. Read alongside customer lifetime value, it turns the acquisition budget into a decision instead of a guess.
Common mistakes with cohort analysis
The most frequent is reading young cohorts as though they were finished. The newest row has had the least time to collect repeat orders, so it will always look worst. Compare like ages, not like dates. The second is cutting cohorts so finely that each one holds a handful of customers, where a single large order swings the whole line — group up until the pattern holds steady.
A third is treating a dip as a marketing failure when it was an operational one. A cohort acquired during a stock shortage, a delivery backlog or a payment gateway outage will retain badly for reasons no campaign caused. Keep a note of what was happening when each cohort arrived, or the table will mislead you later.
How to act on it
Start with one table — first purchase month down the side, repeat revenue across — and rebuild it every month rather than once a year. Then split it by acquisition channel, because the channel that looks cheapest on the first order is often not the one producing customers who return.
Use what you find on the spending side. Move budget towards channels whose cohorts keep paying, and treat a cohort that never repeats as a signal about the product, the offer or the after-sale experience rather than about the ad. Trustworthy cohorts need clean order data with a stable customer identifier, which makes this an analytics and tracking job before it is an analysis one.