Cohort analysis groups your customers by some shared starting characteristic — most commonly the month they made their first purchase or signed up — and then tracks each group's behaviour separately over the months that follow. Cohort 1 (January signups) gets traced through their second, third, twelfth month; Cohort 2 (February) the same. You compare cohorts head-to-head at the same age.
Why it matters
Aggregate metrics lie because they conflate cohorts. If your retention rate this month looks better than last month, it might be that this month's older customers are pulling up the average — not that your retention actually improved. Cohort analysis isolates the effect: what's the 90-day retention rate for January's cohort vs February's? Now you can see whether the product, the offer, or the acquisition channel actually got better.
What to track per cohort
- Retention curve: % of cohort still active at month 1, 2, 3, 6, 12
- Cumulative revenue per customer at each month — your LTV curve, segmented
- Repeat purchase rate by month
- Refund / chargeback rate (some cohorts have systematically worse customers — usually those acquired via aggressive promo)
Common misuses
- Treating a single quarter's cohort as predictive when retention plays out over 12+ months
- Cohorting only by date, not by acquisition source — the cohort matters most when sliced by 'how we got them'
- Comparing cohorts at different ages — January at month 6 vs February at month 4 isn't apples-to-apples