Cohort retention analysis. The cut that found the industrials.

The cut that actually mattered

When we did slice the base, two dimensions stood out: industry, and account size — the fish again, minnows through whales. Account size we already understood. Industry was the one that surprised us, and it was the most insightful cut we ever ran.

Not all industries churned at the same rate. Our best-retaining segment turned out to be the industrials — blue-collar companies with less sophisticated sales and marketing engines and more antiquated tech stacks. They churned lower than the rest of the base.

That second sentence is my read on the why; what we knew for certain was the what: industrials stayed. And it's exactly the kind of finding a blended number buries. Average the industrials in with everyone else and you get an unremarkable churn rate that tells you nothing about where your durable revenue actually lives. The cohort cut is what makes the strong segment visible instead of averaged away.

What we did once we saw it

A finding you don't act on is just trivia, so here's the part that justifies the analysis. Once we learned the industrials were our best segment, we pivoted toward them: marketing relationships, keywords, and the tradeshows we chose to attend all shifted to chase more of that buyer.

That's the through-line the two-churns piece pointed at — it ended on the question of which customers stay, and this is where that question gets answered and turned into a decision. Cohort analysis isn't a reporting exercise. It's an ICP-sharpening exercise. The output isn't a chart; it's a change in where you spend your next acquisition dollar.

The cut that found nothing — and why that helped

Here's the cut we wanted to work and it didn't: acquisition channel. We were always trying to assess and optimize it — does the channel a customer came through predict how long they stay? We spent on AdWords, tradeshows, and other lead channels, and tested just about everything. The truth is the patterns were never that clear. No channel reliably produced repeatable, consistent medium-to-large accounts with high retention.

The one unsurprising exception: word of mouth tended to produce the best-quality leads. Beyond that, channel just wasn't a predictor for us.

That sounds like a dead end, but a cohort cut that shows no pattern is still a finding. It told us to stop looking for a retention signal in channel data and put that optimization energy somewhere it would actually move — toward the industry segment that did predict retention. A null result redirects effort; the only waste is not running the cut and continuing to assume the lever works.

Time-based cohorts

Separate from slicing by segment, there's slicing by when customers signed up — does this year's cohort survive to month six better than last year's? We did time-based cohorts only occasionally, not as much as we should have, and I'd put it on the list of things to do more of. Done periodically, it's how you'd know whether an onboarding change or a product improvement actually made later cohorts stickier, rather than just hoping it did. It turns "we improved onboarding" into a curve you can point at.

The minimum version worth running

For a founder at $1–10M wondering whether this is over-engineering on a few thousand small accounts: the minimum useful version is one cut. Run retention by industry and subsegment, and find which ones drive the highest LTV and the lowest churn. That single view is enough to start fine-tuning your ICP, and it's the one I'd tell you to run first.

That's the whole discipline, and it scales down cleanly to SMB. You don't need a cohort dashboard refreshing nightly. You need to run the segment cut a few times a year, act on what it shows the way we acted on the industrials, and resist the urge to over-build the analysis before you've used the simplest version of it to change a decision.