Cohort curves are one of the most seductive charts in e-commerce analytics, and one of the easiest to over-read. At small scale especially, people pull a cohort report, see a line bending nicely upward, and build a CAC ceiling on top of it that the data cannot actually support. Here is how I try to read these without fooling myself.
What a cohort curve actually shows
A cohort curve groups customers by when they first bought, usually by month, then tracks cumulative revenue per customer as each group ages. The January cohort, the February cohort, and so on, each followed across the months since acquisition. Done right it shows you how much a customer is worth over time and how quickly that value accumulates. Done carelessly it shows you mostly noise dressed up as insight.
Small samples lie
The first problem at small scale is sample size. A cohort of eighty customers can be dominated by two or three big spenders, and average revenue per customer will swing wildly month to month for reasons that have nothing to do with your marketing or product. Look at the median alongside the mean, or a trimmed average that drops the extremes, before you trust any per-customer number. If the mean sits far above the median, a handful of whales are running your chart and the average is not describing a typical customer at all.
Do not compare cohorts of different ages
The second trap is comparing a mature cohort to a young one on absolute value. Of course your cohort from a year ago has higher cumulative revenue per customer than the one from last month, it has had a year to make repeat purchases while the new one has had a few weeks. That is not a decline in customer quality, it is just truncation. Compare cohorts at the same age, January at month three against June at month three, or you will invent trends that are pure artifacts of the calendar.
Read the shape, not just the headline number
The most useful information in a cohort curve is the shape, not the final LTV figure. How quickly customers come back for a second order. What share ever place a second order at all. Whether the curve keeps climbing or flattens after the first ninety days. Two brands with the same twelve-month LTV can have completely different businesses, one with high repeat rates and a curve that keeps climbing, one front-loaded with a single purchase and a flat line after. The shape tells you whether you have a retention business or an acquisition treadmill.
Do not spend against LTV you have not earned yet
This is the one that actually costs people money. It is tempting to take a projected long-run LTV, sometimes extrapolated from thin data, and use it to justify a high CAC today. The problem is you are spending real money now against revenue that is a forecast, and if the forecast is optimistic you are buying customers at a loss for months before you find out. I prefer to set spend ceilings against a conservative realized window, something like ninety-day realized contribution that has actually happened, and treat longer-run LTV as upside rather than budget. If the long tail shows up, good, but you have not bet the business on it.
Watch for seasonality and source
A cohort acquired during a holiday promo or a viral moment behaves differently from one acquired in a quiet month through steady spend. Discount-driven cohorts often have weaker repeat behavior. If you blend all acquisition sources and seasons together, you get an average that describes none of them. Segment by acquisition month and, where you can, by source, so you are comparing like with like.
A cohort curve is a hypothesis generator, not a forecast you can take to the bank. It is great for spotting that something changed, that a recent cohort is repeating faster, that a channel brings worse retention, and worth almost nothing as a precise prediction at small volume. Read it for direction and shape, stay skeptical of the headline number, and never let an unearned LTV write checks your bank account has to cash.
Leave a Reply