Purchase frequency is how many times the average customer bought from you in a set period.
Nine of the thirteen publishers read for this page divide total orders by unique customers over one window. Real agreement across this source set.
Underneath it, four things move. The unit. The population in the denominator. The window. And the arithmetic that turns a count into an interval.
Same formula. Different number.
Purchase frequency, as the publishers document it. The ratio is widely shared, and it is not enough to make two numbers comparable. One guide gives the name to a duration in days; one platform ships days and a predicted order count. The circulating benchmark was computed on repeat buyers alone, while the formula printed above it on two pages here counts every buyer.
What is purchase frequency?
It is the orders a store took over a period, divided by the different customers who placed them.
LoyaltyLion writes it as “Purchase Frequency = Total Orders in Period / Total Unique Customers in Period”.
SurveyMonkey writes the same ratio as “Customer purchase frequency = Number of orders ÷ Number of unique customers”.
The first works it through: “suppose over the last quarter you had 500 orders placed by 350 unique customers. Purchase frequency = 500 / 350 ≈ 1.43”.
What else is this number called?
One glossary lists the aliases in one line: “Also known as: order frequency, transaction frequency, buying frequency, average purchase frequency”.
One guide names a second collision: “It is important to distinguish between customer purchase frequency, which measures actual transactions, and ad frequency, which tracks impressions per user.”
Both divide a count of events by the people those events touched.
Is it a count, or a number of days?
Nine publishers return a count. One returns a duration and gives it this metric’s name.
Kapiche’s retention guide writes that “Time Between Purchases, also known as Purchase Frequency or Inter-Purchase Time, measures the average duration between consecutive purchases made by the same customer.”
One platform ships the duration as a field. Klaviyo’s predictive analytics publish “Average Time Between Orders” — “The average number of days between each of a customer’s orders.” — with a profile value of “75 days”. The same page also discusses a predicted number of orders — a count, not a duration.
A third vendor keeps both apart: one metric page defines “purchase frequency = total orders/# of unique customers”, and “Time Between Purchases” is another.
Which customers belong in the denominator?
Every customer who placed an order in the window, on the reading nine publishers share.
One glossary states it flatly: “The key here is tracking unique customers, not the total number of customers.”
Another says it as a threshold. The denominator is “the count of distinct customers who placed at least 1 order during the same period”.
The circulating benchmark does not use that population. The study behind it counts only buyers who came back: “In our study, a loyal customer is taken as any customer who buys from your store more than once.”
One worked example does the same, dividing “Total number of purchases made by these customers: 35,000” by “Total number of customers who made at least two purchases: 5,000”.
One guide shortens the denominator to “Total customers” in its displayed formula, after instructing readers to divide by “total unique customers in the same period”; its example uses 400 unique customers.
How much does that choice move the number?
Not from any single store’s figures. On the closest worked example here, a repeat-only ratio cannot be calculated.
Its illustration is a pottery shop that “processed 50 orders in Q2 from 35 unique customers”. Of those, “25 customers each made a single purchase, while 10 customers accounted for the remaining 25 orders”. The page prints the all-customer answer: 1.43.
Those 10 averaged 2.5 orders each. How many bought more than once, it never says.
Why might your number not match the benchmark you read?
Because on two of these pages the formula and the benchmark count different people, and neither says so.
The glossary that tells you to track every unique customer then prints category figures lifted from the loyal-customer study: “A study by customer retention platform Beans found the following benchmarks for eCommerce brands by category:”
The guide that prints “Purchase Frequency = Total Orders in Period / Total Unique Customers in Period” near the top prints, further down, “loyal shoppers (those who make more than one purchase) averaged about 3.58 purchases per year.”
How big is the difference?
About two to two and a half times — but only on an assumption the study does not make: that its loyalty shares and its 3.58 figure describe the same customers in the same year, and that everyone else bought once. That is an effect size, not a store benchmark.
Its loyal customers average 3.58 purchases a year. Its loyal-customer shares run from 29% in consumables to 16% in Fashion and General Goods.
Read the 3.58 as repeat buyers and give everyone else a single order. The all-customer ratio lands at 1.75 at the top share and 1.41 at the bottom — the printed 3.58 divided by 2.05 or by 2.54.
Where do the circulating benchmarks come from?
Two pages here reprint numbers from the same Beans study.
It states its own scope, and its own limits. “The sample includes 495 retailers”, 2.6 million orders from 1.6 million customers, “representative of small and medium-sized online shops”. Five named categories hold “less than 10 stores”, industry figures are “weighted averages”, and it saw only store-account customers.
It states its window — “a period of 1 year” — and its purpose: “To create reliable and up-to-date purchase frequency benchmarks for e-commerce retailers to use in 2017.”
Its headline: “customers buy between 3 and 4 times per year regardless of the industry they’re buying in”.
Do the published ranges agree with each other?
No. Across the bands published here, consumables and food run from 3.12 a year to an annualised threshold of 24 or more — a spread of at least 7.69 times.
One glossary prints “Benchmark: Consumables: 6-12 purchases/year; fashion: 3-6 purchases/year; electronics: 1-2 purchases/year” and cites nothing for it.
A metric dictionary prints four bands in two different units — “4–6+ purchases per customer per quarter” for grocery and consumables, against “3.5+ purchases per customer per year” on the row it heads top-performing, printed below it. Annualised, that quarterly row starts at 16 a year — 4.6 times the 3.5 minimum. Both plus signs leave the top of their band unstated.
The reprinted study puts electronics at 4.17, against 1-2 a year in the first table.
None of these pages states which denominator its band was built on.
What window should you measure over?
Twelve months, on three of these pages, unless the category buys faster than that.
One KPI page says “it’s most commonly looked at over a 12 month period to take seasonality and promotions into account”. A glossary says “Use a 12-month window unless your product has a shorter natural buying cycle.”
Two publishers put a floor under the window as well: “For most stores looking at Purchase Frequency for less than a quarter won’t make sense.”
The metric dictionary makes it a category question instead. “A monthly view suits high-frequency categories like grocery or coffee. A quarterly or annual view is more appropriate for apparel, electronics, or home goods.”
Does a longer window make the number look better?
Two publishers say it does — one flatly, the other as a tendency. Numbers from different windows do not compare.
“The longer the time frame used, the higher your Purchase Frequency will be.” And: “note that the longer the time frame that you use, the higher your purchase frequency will tend to be”.
They also state the floor the ratio cannot drop below inside one window: “Your Purchase Frequency will always be one or greater if you’re using the same period for the number of unique customers as the number of orders.”
One page calls it “a laggy metric that can only be looked at for longer periods of time”, because each customer needs time to buy twice.
What counts as an order?
That is a decision, and one source here says so out loud.
The metric dictionary tells you to “Decide upfront whether to count gross orders or net orders (after returns).” Its own recommendation: “Net is more accurate for understanding true purchase behaviour, but be consistent across periods.”
Shopify’s analytics reference documents the default it is made against. Returns land in the order count unless you filter them out: “Returns generate order records to maintain accurate inventory and financial tracking.”
The same dictionary names the other direction too: “In some retail contexts, splitting a single purchase into multiple transactions inflates frequency.”
Who counts as one customer?
Whoever your records can tell apart: three publishers reach for an email address, an IP or physical address, and a store account.
One tracks an email address in a CRM: “Even when they use guest checkout, you can add customers to a customer relationship management (CRM) database each time they make a purchase, so you can track unique customers by their email address.”
The benchmark study could see only account holders, and says which way that bends its own customer count: “We were only able to access data related to customers who had store accounts which means the number of total customers is likely to be higher than stated.”
Duplicates move it in a known direction: “If your system counts the same customer under multiple IDs, unique customer counts inflate and frequency drops artificially.”
Why do new customers pull the average down?
Because they have had less time to buy again. Two publishers warn that new customers drag down the average; one recommends separating cohorts by join date.
One lists it as a common mistake: “Not accounting for customer tenure - including brand-new customers in the average drags down the frequency number”.
The other states it as a property of the number: “New customers have lower frequency by definition.”
Its remedy: “Separating cohorts by join date prevents newer customers from dragging down your overall average.”
Does a store platform compute this for you?
One does, under a different name and a narrower scope.
Shopify’s field reference carries no field called purchase frequency. Its ratio is “Orders per customer”, described as the “Average number of orders placed by customers in the given cohort”.
A cohort there is customers who share the “Month the customer placed their first order” — neither every buyer in your window nor the repeat-only subset.
The per-customer count it does ship carries no window at all: “Number of orders the customer has placed across all time”.
How do you turn the count into an interval?
Divide 365 by it, when the count is an annual one. One glossary does exactly that: “12,000 orders from 5,000 unique customers over a year gives a purchase frequency of 2.4 orders per customer per year, or one order roughly every 152 days.”
The benchmark study runs the same arithmetic from the other end: “a store with loyal customers who buy on average every 73 days would return a purchase frequency of 5 (365/73)”.
The study’s benchmarks above, then, were produced from average intervals between purchases, not by dividing orders by customers.
Why do two conversions of one count disagree?
Because one rule counts the purchases and the other counts the gaps between them.
A retention guide divides by the count less one: “Average Time Between Purchases = 365 / 7-1 = 60.83 days”. The study divides by the count itself.
Apply the study’s rule to that guide’s own worked store and 365 ÷ 60.83 is 6, against the 7 purchases it started from.
Take the published count of 2.4. One rule returns 152 days; the other returns 261.
Read “365 / 7-1” left to right and it is 51.14, not the 60.83 printed beside it. The rule is recoverable from the answer, not the expression.
Is an average the right statistic here?
The ratio is an arithmetic mean. Use it as a headline, but not alone: one-time buyers and a few very frequent buyers pull it around, so one publisher also recommends the median interval among repeat customers.
“The mean is pulled around by one-time buyers and by a small number of very frequent ones”, the same page writes. Its advice is to “also calculate the median time between consecutive purchases for repeat customers.”
No source here publishes one for a real store.
How does this feed customer lifetime value?
Two publishers put it alongside average order value and customer lifespan in calculating lifetime value.
One writes it as “CLV = average order value × purchase frequency × customer lifespan.” The other names the same three factors. The customer lifetime value page covers where those formulas disagree.
What travels badly is the unit. In one worked chain the order-value step is computed on a month — “Say you earned $10,000 from 50 orders last month. Your AOV is $200.” — and the frequency step on a year: “If 600 orders came from 400 unique customers in a year, your frequency is 1.5.”
The product is printed as “$200 × 1.5 = $300 per customer per year.” and then multiplied by a lifespan in years. That works only because the frequency step used a year. Had it used the month the step above it used, the same product would price five months.
Is raising it always worth it?
No. Higher purchase frequency can raise revenue, but profitability still depends on margins, cost structure and acquisition expense; heavy promotions can erode margin faster than revenue grows.
One publisher writes that the metric “is unique because it compounds: a customer who buys twice a year at $50 generates $100 annually, but if you increase their frequency to four times a year, revenue doubles to $200”.
Another prices the same move at store scale, then qualifies it: “Higher purchase frequency can generate more revenue, but profitability depends on margins, cost structure, and customer acquisition expenses.”
What should you check before comparing your number to anyone else’s?
Check four things: denominator population; whether the unit is a count or interval; the window used on both sides; and whether returns or duplicate customer records are included.
Which customers are in the denominator. Every buyer in the window, repeat buyers only, or another scoped population such as an acquisition cohort. That is the gap above, and neither of those two pages labels it.
What unit the number is in. A count and an interval both travel under this name, and the two published conversions between them differ by one purchase.
How long the window is, on both sides. Window length moves the number, and the worked examples here run over a month, a quarter and a year.
Whether returns and duplicate records are in. One inflates the numerator by default, the other inflates the denominator.
Where does that leave the metric?
Worth computing, and worth writing down what you computed.
The ratio is widely shared, and it is not enough to make two numbers comparable. The population, the window, the unit and the identity rule all move it — and the circulating benchmark was built on a narrower population than the formula sitting above it.
So compute it the way nine publishers describe, record those four choices beside it, and compare it to your own last year, not a category band.
Here, the denominator is the argument.