Cohort (Cohort analysis)

Worked example (made-up numbers)

Customers who started paying in July: 20. Still paying three months later: 12. Customers who started paying in August, after you rewrote onboarding: 25. Still paying three months later: 19.

In total, you have more customers than ever, which says little. Split into cohorts, the answer is clear: 60% of the July cohort stayed three months, 76% of the August cohort did. The onboarding change is working.

Why totals mislead

A total mixes old and new users. New signups can hide the fact that everyone who joined three months ago has left, and a burst of early customers can make a worse recent month look fine. Cohorts compare like with like: people at the same age with your product.

Cohorts by more than date

The same idea works for any shared starting point:

• By source: customers who came from search against those from a sponsorship.
• By plan: monthly against yearly.
• By offer: those who joined with a launch discount against those who paid full price.

Grouping by source is often the most useful early on: it shows which channels bring people who stay.

Why it matters when you’re starting out

Early numbers are small and noisy. Cohorts are how you tell whether a change helped: compare the group that started before it with the group after. Without them, every month’s total goes up a little and you can’t tell whether you’re improving or just adding.

How to do it with a spreadsheet

1. One row per customer: start month, source, and whether they’re still active.
2. Group rows by start month.
3. For each group, count how many are still active after one, two and three months.
4. Put the groups side by side.

In Nordva Launch

Launch doesn’t draw cohort tables. It gives you the columns for one: each paying customer’s first-visit source, and from Stripe (read-only, every plan) when they started and whether they cancelled. On Indie and up, the contacts list puts both on one record per person, which is enough for the spreadsheet above.