Repeat purchase rate: which formula, and why the same data gives different answers

Repeat purchase rate is the share of customers who bought more than once in a period: customers with two or more orders divided by all customers. That is the customer-based version. An email marketing software vendor's glossary gives the same formula, repeat customers divided by total customers, "within a given time period"². A loyalty and marketing software vendor's guide describes it as the percentage of customers who make more than one purchase, without naming a window³. The analytics vendor above also cites an order-level version: purchases from repeat customers divided by all purchases in the period¹.

Three versions turn up in practice:

1. Customer-based rate. Customers with two or more orders in the window, divided by all customers who ordered in the window.
2. Order-based rate. Repeat orders, meaning orders from customers who had bought before, divided by all orders in the window.
3. Cohort-based rate. Of the customers whose first order fell in one month, the share who bought again within a fixed time afterwards.

The first answers "how many of my customers are repeaters?". The second answers "how much of my order volume comes from repeaters?". The third answers "do customers acquired in a given month come back, and how fast?". Different questions, different numbers.

Three choices move the number. Each is legitimate. Mixing them is what breaks comparisons. The analytics vendor traces its gap to two of them: whether the denominator is customers or orders, and whether a repeat buyer's first order counts¹.

Customers or orders. A small group of frequent buyers can make up a large share of orders and a small share of customers. The customer-based rate stays low while the order-based rate looks high.

The window. The same vendor puts it bluntly: a repeat purchase rate is meaningless without a window¹. A rate measured over twelve months counts more repeaters than one measured over three. A customer who bought in January and again in November is a repeater in one and not in the other.

The population. Do you count all customers who ordered in the window, only new customers, or everyone ever? Counting old customers who bought again lifts the rate. Counting only new customers in a short window lowers it.

Example (illustrative): A shop has 1,000 customers and 1,500 orders in a year. 200 customers ordered twice or more, placing 700 orders between them. Customer-based: 200 ÷ 1,000 = 20%. Order-based: 1,000 orders are first orders, so 500 are repeat orders, and 500 ÷ 1,500 is about 33%. Count every order from repeat buyers, first orders included: 700 ÷ 1,500, about 47%. Same shop, same year, three honest answers.

The first rule of reporting: state the formula and the window next to the number.

Retention is the share of customers who buy again within a given period, usually tracked by cohort. Churn is the rate at which you lose customers: customers lost divided by customers at the start. For a subscription business, a customer is gone when they cancel.

Shops have no cancel button. One analytics vendor argues that the standard retention formula was built for subscription businesses with contracts, breaks in stores without them, and should be replaced by the share of each first-order cohort that places a second order within a window⁴. A commerce platform's blog counts a non-subscription customer as churned when they have not reordered within twice the average time between repeat purchases⁵. That multiplier is one platform's rule of thumb. Our view: both points are right in principle, and the window should come from your own data.

So you must define "gone" yourself. A customer who orders coffee monthly has probably churned after several missed months. A customer buying a sofa may not return for years. Set the window from your own order gaps, then write it down.

A product analytics vendor's guide shows the contrast. It gives the subscription-style formula, customers at the end minus new customers, divided by customers at the start, and its examples are mostly apps⁶. The same guide reads retention from cohort charts, where each row is a group of users who started at the same time⁶. For a shop, the cohort part is the bit worth keeping.

Cohort analysis groups customers by a shared trait, usually the month of their first purchase, and tracks their behaviour over time. It fixes the population problem. You compare like with like: customers acquired in March against customers acquired in April, each measured at the same age.

What it shows that a single rate hides:

• Whether newer cohorts repeat more or less than older ones.
• How quickly customers come back after the first order.
• Whether a change in acquisition, such as a new channel or promotion, brought different customers.

Google Analytics 4 has a retention overview report that groups users into cohorts by the day they were acquired⁷. It counts users who come back to the site or app, not customers who buy again⁷. Our note on GA4 cohort exploration covers what it can and cannot show. Treat GA4 as a starting point; your order data is the source of record.

Cohorts also feed LTV (customer lifetime value), which is what a customer is worth over time. Sources use three approaches: historical, predictive and cohort-based. Repeat behaviour by cohort is one of the inputs for the last two.

Our view: for a consumer ecommerce brand, report two numbers and one table.

1. Customer-based repeat purchase rate over a fixed trailing window, such as twelve months. Pick the window from your buying cycle. Do not change it every quarter.
2. Order-based repeat share, labelled as such, so nobody confuses it with the first number.
3. A cohort table: first-order month down the side, time since first order across the top, share who bought again in each cell.

And keep these rules:

• Write the formula and window under every chart.
• Leave out cohorts that are younger than your window. The analytics vendor above advises the same¹.
• Never compare your figure with a benchmark whose definition you cannot see.
• Keep the definition fixed for at least a year, so trends mean something.
• Link repeat behaviour to margin. A higher repeat rate that costs a deep discount is not progress.