Definition
Cohort analysis reveals the truth about your business — whether each new customer cohort is retaining better or worse than previous ones. GitDealFlow's weekly signal updates follow a similar cohort pattern: tracking engineering momentum over time in weekly buckets.
How it works in practice
Cohort analysis groups customers by a shared starting event — typically the month they signed up — and tracks each group separately over time. Aggregates hide the truth: a flat overall retention number can hide a product that is getting better for new customers while old cohorts decay. Cohorts make those trends visible.
The standard output is a retention curve per cohort: the percentage of each cohort still active in month 1, 2, 3, and so on. Reading the curves matters more than any single number — are newer cohorts retaining better than older ones? Does retention flatten after an initial drop (the classic SaaS usage curve), or keep sliding? The first pattern means the product is improving; the second means the fixes are not working.
Cohort thinking extends beyond revenue: it applies to activation, engagement, and even team behavior. GitDealFlow publishes its engineering data in weekly buckets — tracking commit velocity, contributor growth, and repository expansion across 4,200+ startups — so investors can compare how a company's momentum is trending cohort over cohort, the same way a SaaS operator reads retention curves. Its SSRN research panel has documented 219 fundraises, with acceleration typically visible 21-47 days before a round is announced.
Key points
- A cohort is a group sharing a start date — usually signup or first payment month.
- Retention curves per cohort reveal whether the product is improving over time.
- Flat aggregates hide offsetting trends between cohorts.
- The shape of the curve — flattening vs. sliding — is the signal.
- Cohort logic applies to engineering momentum too: compare week over week, not just totals.
Frequently Asked Questions
What does cohort analysis tell you?
Whether product, marketing, or sales improvements actually work. If the Jan 2026 cohort retains better than the Oct 2025 cohort, your improvements are real. If not, you have a fundamental problem.
What is a cohort?
A group of users or customers who share the same starting event, usually the month they first signed up or paid. Tracking cohorts separately — instead of averaging everyone together — lets you compare how customers acquired in different periods behave, which isolates the effect of product changes from the effect of acquisition timing.
Why is cohort analysis better than aggregate metrics?
Because averages can cancel out. Overall churn might look stable at 4% while new cohorts are actually churning at 8% and old ones at 2%. Cohort analysis reveals which direction the business is really moving — whether the latest cohorts are better or worse than earlier ones — and points to what caused the change.
What does a good retention curve look like?
Typically a sharp initial drop as casual users leave, then a flattening as the retained base stabilizes — the classic usage curve. A curve that keeps declining month after month, without flattening, means the product is not retaining anyone long-term. The key comparison is between cohorts: each new cohort should flatten at a higher level than the last.