Repeat purchase rate
The metric that decides whether you have a business or a customer acquisition habit, and the one most often computed on a window too short to show it.
Also called Repeat customer rate, Returning customer rate.
What it is
The share of customers who have bought more than once. It is the closest thing to a single measure of whether the product is worth buying again, and it moves slowly enough that a single month tells you nothing and a six-month trend tells you a great deal.
customers with 2 or more orders ÷ customers with 1 or more orders
- customers
- Distinct people, after resolving duplicate records
- orders
- Completed orders within the measurement window
The part the formula leaves out
The window decides the answer. A repeat rate measured over three months excludes every customer whose natural repurchase cycle is longer than three months, which for most considered purchases is most of them. The number is not wrong; it is answering a question about three months while being read as a question about the business.
Worse, the window truncates at both ends. Customers acquired in the last month have not had time to come back, so including them in the denominator guarantees the rate falls whenever acquisition rises. A store having its best-ever month for new customers will show its repeat rate collapsing, and the two facts are the same fact.
Then identity. Two orders from the same person under two email addresses are two customers who each bought once. Deduplication is not tidiness here, it is the measurement: in the worked donor example, resolving people rather than records moves retention by eleven points, and the same arithmetic applies to any customer file with a web form attached to it.
How it is usually computed wrongly
01
Including recently acquired customers in the denominator
Someone who bought last week has not had the opportunity to buy again. Counting them as a non-repeater makes the rate a function of how fast you are growing.
Use cohorts. Measure each acquisition month's repeat rate at a fixed age, so what you are comparing is the same opportunity to return.
02
Counting records rather than people
Guest checkout, a changed email address and a married name each split one customer into two, and every split moves a repeater into two one-time buyers.
Resolve identity on email, and on name with address, before counting. Report how many records collapsed, because that number is itself the size of the correction.
03
Measuring over a window shorter than the repurchase cycle
A product bought once a year has a repeat rate of nearly zero over any quarter, and the quarterly figure will be read as a retention problem that does not exist.
Set the window from the observed distribution of gaps between orders, not from the reporting calendar. If the median gap is 140 days, a quarterly window cannot see it.
04
Reading it as a percentage of orders
Share of orders from returning customers and share of customers who returned are different metrics that both get called repeat rate, and the first is always higher because repeaters order more.
Say which one you mean every time. Both are useful and they answer different questions: one is about revenue mix, the other is about the product.
What your file needs
- A customer identifier, or the fields needed to construct one
- One row per order, with a date
- Enough history to cover at least two repurchase cycles
Anything missing is reported as unavailable rather than substituted with something weaker computed on worse evidence.
Compute it on your own file
No account needed to start. You only pay when you like what you see.