What is Cohort Analysis?
Grouping customers by when they first bought, then tracking how each group behaves over time.
A worked example
Customers acquired in January are tracked separately from those acquired in February. By month three, the January group has bought 1.8 times on average and the February group 1.1. The difference points at something that changed in acquisition, most often a new channel bringing worse customers at a better cost per order.
Why it matters
Cohort analysis is how you find out whether your business is actually improving, because blended averages hide decline. A brand growing quickly will show a healthy overall repeat rate purely because new customers keep arriving, right up until the moment growth slows and the underlying deterioration becomes visible.
It is also the only honest way to build lifetime value. Projecting LTV from a blended average assumes every customer behaves like the average, which they do not. Watching real cohorts mature tells you what customers acquired through a given channel are genuinely worth, and it usually differs by channel more than brands expect.
For anyone spending on acquisition, this is the difference between knowing your economics and estimating them. Cohorts show whether the customers you bought last quarter are paying you back on the schedule your model assumed.
The nuance most people miss
A cohort younger than your purchase cycle tells you nothing yet, and reading it anyway produces confident wrong conclusions. If customers typically reorder at ninety days, a thirty-day-old cohort has not had the chance to repeat, and treating its low repeat rate as a signal will send you chasing a problem that does not exist. Let cohorts mature past at least one full purchase cycle before drawing conclusions, and compare cohorts at the same age rather than at the same date.
Common mistakes
- Comparing cohorts at the same calendar date rather than at the same age
- Reading a cohort younger than one purchase cycle and acting on it
- Blending all cohorts into one average, which hides deterioration during growth
- Building lifetime value from a projection when observed cohort data exists