The retention number measured on the wrong cohort

At one company, the Series B deck said 92% net retention. The number was accurate as a blended figure across the entire customer base. It was also misleading. The two-year customers, the early adopters who'd been through the product's strongest period, were retaining above 100% net at that company, because expansion inside those accounts ran heavy. At that same company, the customers who'd joined in the last year were retaining at 68%. The blend averaged the two into a figure that looked healthy and buried the trend.

This was a Series B deck, and the investor's diligence team ran the cohort analysis. The recent cohorts looked bad enough that the term sheet came back with retention concerns flagged. The founder explained the blend. The investor adjusted the valuation. The round closed below the price the deck had implied, because the deck told a story the cohorts didn't support. A blended number can say something true and mislead anyway.

Blended retention is one of the most common metrics in SaaS reporting and one of the most consistently misleading. The blend averages across customers with different tenure, different product experiences, and different reasons for staying. Early customers tend to retain better because they were selected for fit when the product was less mature. They were enthusiastic about the mission, willing to tolerate rough edges, often builders themselves. Recent customers arrived through more polished marketing, into a more crowded field, and they retain worse because the bar for staying has risen and the product hasn't risen with it.

The blend hides this drift in both directions. A company with declining cohort retention can post stable blended retention for two to three years, because the older cohorts keep performing and dominate the average. By the time the blended number drops, the trend has been visible in the cohort curves for many quarters, and the company has already missed chances to act. The blended number is the lagging indicator. The cohort curve is the leading one. It's the same failure behind the churn that was legible long before anyone named it.

The discipline is to report both. The blend is fine as a headline for investor decks and board updates. Below the headline, the cohort curves belong on the page every month, by signup quarter. The most recent cohort gets the most attention. If it's retaining worse than the cohort before it, the company has a trend, and investigation follows.

The setup is straightforward. Group customers by signup month or quarter. For each group, track the share still subscribed at month one, three, six, twelve, twenty-four. Plot the curves on one chart. Most data tools support this natively and the SQL runs a few lines. An analyst can build it in a day. That most companies don't isn't a technical limit. It's a cultural one. The blend is easier to celebrate, and the cohort curves usually tell a more complicated story.

The most expensive version of this is the founder reading their own blended retention as healthy and concluding the business is healthy. The recent-cohort decline is invisible to them. Six months later the blend finally moves, and they're now answering to a problem that's been in the data for a year.

Report the blend. Watch the cohorts. A company is not its average customer. It's the customer it's signing this quarter, and that customer is on a curve of their own.