Customer cohorts
One retention percentage tells you how many of your customers have ordered twice. It cannot tell you whether the ones you won this quarter are behaving better or worse than the ones you won last year. Cohorts can.
What a cohort is here
A cohort is everyone whose first order fell in the same quarter. Someone who first ordered in July sits in the Q3 cohort and stays there for good, however long they go on ordering.
It is keyed on the first order, not on when the customer record was made. An operator can add Maya Reyes at the counter in March and she may not buy anything until August — she belongs to the quarter she first ordered in, not the quarter somebody typed her name. A cancelled order does not start a cohort either; work that was called off is not an acquisition.
The columns
| Column | What it counts |
|---|---|
| Cohort | The quarter of the first order, for example Q3 2026 |
| Customers | How many people first ordered in that quarter |
| Ordered again | The share of them that have ordered more than once, ever since |
| Within 90d | The share that ordered again inside 90 days of their first order |
| Orders each | Total orders from the cohort, divided by its size |
| Value each | Total revenue from the cohort, divided by its size |
Orders each and value each cover the cohort's whole life, not only the quarter it was acquired in. A cohort won two years ago has had two years to spend, and the column says so.
Reading it fairly
A cohort acquired last month will show a lower "ordered again" than one from two years ago, and that is arithmetic rather than a problem — it has had weeks rather than years. The within-90-days column is the honest early comparison, because every cohort old enough to appear in it has been given the same 90 days.
The period picker selects cohorts by when they were acquired. Narrowing to the last 6 months does not trim a cohort's history; it removes whole cohorts whose first orders fell outside that span, and leaves the surviving ones with all of their later orders intact.
Under the table, a line adds the cohorts together: how many customers were acquired in the period, and what share of them have ordered more than once.
What is not in it
There is no satisfaction score on this report. Ratings are collected, but they live on each customer's own record and nothing here aggregates them. The report says so in a line under the table rather than quietly leaving a gap, and no number on the page stands in for it.