Cohort

A group of users who share a starting event in the same period, for example everyone whose first purchase was in September. Cohort analysis follows each group over time.

Formulacohort = everyone whose first [event] fell in the same period

How we work it through

1
Define the starting event: registration or first paid order.
2
Check where the data begins: the first cohort collects everyone who bought before it.
3
Compare cohorts at the same age.

On the practice dataset

The query we run on Meridian, Atterna’s practice marketplace, and what it returns.

SELECT date_trunc('week', first_paid) AS cohort, COUNT(*)
FROM (SELECT buyer_id, MIN(created_at) AS first_paid
      FROM meridian.orders WHERE status = 'paid'
      GROUP BY buyer_id)
GROUP BY 1 ORDER BY 1;

Result1,710 · 1,297 · 494 · 151 — the first week is big because the data starts there, the last has one night

The common trap

Mixing cohorts of different ages in one average, which makes a stable product look like it is declining.

When an AI answers

An AI assistant

“New customers fell from 1,710 to 494 a week.”

What it didn’t know: It did not know the extract starts on 31 August, so “first order” means “first order we can see”.
Ask back: “When does the data start, and is the first cohort really new customers?”

Paste a real AI answer into the checker — it writes these questions for you →