Purchase frequency is the average number of times each buyer purchases a brand, a product or a category within a set period. It is calculated by dividing the number of purchases by the number of people who bought at least once. The measure exists because any sales figure can be split into two questions: how many people bought, and how often each of them came back. Andrew Ehrenberg, the statistician whose consumer panel work defined the field, called average purchase frequency per buyer "the most basic measure of repeat-buying in the theory".

How the number is built

In Ehrenberg's notation, penetration, written b, is the proportion of people buying an item at all in a period; purchase frequency, written w, is the average number of times those buyers buy it. Sales, in his formulation in Repeat Buying, "equal the number of buyers times the average number of purchases per buyer times" the packs bought per occasion and their price. His worked example: 22 of 983 panel households bought Lux soap flakes in 12 weeks, making 29 purchases between them, a penetration of 2.2% and a purchase frequency of 1.3.

The period decides the answer: an annual figure exceeds a quarterly one, but by less than four times, because many buyers recur across quarters.

Online retail uses the same division. Shopify defines purchase frequency as total orders divided by unique customers, and multiplies it by average order value (AOV) and average customer lifespan to estimate customer lifetime value (CLV). A shop taking 12,000 orders from 5,000 customers in a year has a purchase frequency of 2.4; at an AOV of $40 and a three-year lifespan, each customer is worth $288 in revenue. Google Analytics 4 reports the ratio as transactions per purchaser, beside a purchaser rate that plays the role of penetration, according to Google's Data API schema.

Consumer panels publish the pair together. Kantar's Brand Footprint ranking, now produced by Worldpanel by Numerator, counts consumer reach points, the number of times a brand is chosen in a year. Its May 2019 edition showed why both halves matter: Colgate reached 60.5% of households worldwide, which bought it 5.5 times on average, while Indomie reached 4.7% and was chosen 33.1 times per buyer.

Where it enters advertising

Purchase frequency is not a bid request field. It enters upstream, in audience segments, bidding rules and measurement windows, supplied by whoever holds transaction records: retailers, payment companies, receipt apps and panels.

The oldest targeting form is RFM segmentation, which scores customers on recency, frequency and monetary value. The IAB Australia Retail Media Council's December 2025 blueprint lists it as a foundation for lifetime value modelling and retention campaigns. PayPal pitched purchase frequency alongside category preferences and spending behaviour when it opened its transaction audiences to PubMatic buyers in September 2025. Media.net, packaging receipt data from Fetch, which processes more than 13 million receipts a day, argued in April 2026 that in high-frequency categories such as household products the date of the last purchase predicts the next cycle.

Bidding systems use it as a value signal. Google's lifecycle modes, announced in April 2025, include High Value Win-Back for lapsed customers whose past value came from high purchase frequency, large baskets or expensive orders, with returning customers detected from purchases in the previous 540 days.

Measurement windows depend on it too. Amazon Ads treats a shopper as new to brand after no conversion with that brand in 12 months, according to Amazon, a single window for milk and sofas alike. A draft IAB Europe standard released on September 17, 2026 instead ties the lookback to the purchase cycle: up to six weeks for milk or nappies, seven to 26 weeks for jeans or mascara, and 27 weeks or more for televisions and sofas. Amazon Marketing Cloud extended its lookback from 13 to 25 months in November 2025, long enough to observe seasonal repurchase and durable-goods cycles.

Origin and evolution

The modern measure came out of consumer panel research, which Ehrenberg joined in 1955. In March 1959 the Journal of the Royal Statistical Society published his paper The Pattern of Consumer Purchases, showing that purchases of non-durable goods fitted the negative binomial distribution (NBD): most buyers buy rarely, a few buy often, and the shape repeats across products. Repeat-Buying: Theory and Applications followed in 1972, with a second edition in 1988.

Brand choice came next. Goodhardt, Ehrenberg and Chatfield published the Dirichlet model in October 1984, predicting each brand's penetration and purchase frequency from its market share. The sociologist William McPhee had coined double jeopardy in 1963 for a pattern the Dirichlet reproduces. Double Jeopardy Revisited, in the Journal of Marketing in July 1990, put it plainly: a small brand has far fewer buyers, and "its buyers tend to buy it less often".

Customer-base analysis carried the counts to individuals. The Pareto/NBD model of David Schmittlein, Donald Morrison and Richard Colombo, published in 1987, estimated from each customer's frequency and recency whether that customer was still active; Peter Fader, Bruce Hardie and Ka Lok Lee simplified it in 2005.

Byron Sharp of the Ehrenberg-Bass Institute at the University of South Australia turned the regularities into strategy with How Brands Grow in 2010, arguing that because frequency varies little between brands, growth comes chiefly from penetration.

Why the split matters

The practical question is which lever moved. Charles Graham, writing for The Marketing Society in July 2012, cited instant coffee data in which Nescafe held a 45% share with 34% penetration and Birds a 2% share with 2% penetration, yet buyers of any brand bought about twice a quarter against a category average of 3.2. The gap between the brands was almost entirely one of reach. Double jeopardy, he noted, can be written as w(1-b) equals a constant.

Scale confirms the pattern. A 2025 study in the Journal of Business Research by Steven Dunn and seven co-authors decomposed revenue in 474 US packaged-goods categories over 13 years of household panel data into penetration, purchase frequency, volume per trip and price per volume. Penetration and price drove growth. Changes in purchase frequency had minimal impact, and penetration loss was the leading cause of decline.

Commerce media sells the other lever. Amazon said customers who added fresh groceries to Same-Day Delivery orders shopped roughly twice as often as those who did not. Criteo's CPG Pulse 2026 found that Black Friday buyers who returned spent about six times as much as those who did not, with orders only about 6% larger: the difference was frequency, not basket size. Both figures are company-supplied.

Where it breaks down

The loyalty argument is not settled. In January 2018 Dunnhumby, whose analysis drew on Tesco shoppers, argued that heavy buyers contribute more to sales than Sharp's work implied. Sharp replied that "a flawed sample isn't fixed by making it bigger", according to Campaign. Dunnhumby's Adam Smith and Nick Blair countered: "We see many light buyers stop buying altogether, rather than naturally moderate upwards." Sharp, Jenni Romaniuk and Graham later put the heaviest 20% of buyers at not much more than half of sales, a 60/20 rule rather than 80/20.

Averages conceal the distribution. Under the NBD, a frequency of 2.4 can rest on a base where most buyers bought once and a small group bought many times, so a rising average may reflect a handful of heavy buyers.

Counting customers is harder than counting orders. Guest checkouts split one buyer into several, depressing frequency, and Google's schema lists subscription renewals and refunds among GA4 transaction events. A retailer's figure also sees only its own till: a shopper who buys the brand elsewhere looks infrequent. In non-contractual retail, a lapsed buyer and a slow one look identical, so frequency-based retention models infer departures rather than observe them.

RFM has its own caveat: the IAB Australia blueprint called it static and backward-looking, liable to overvalue one-off bulk buyers. Predictive versions promise more and disclose less. Sam's Club said in September 2026 that it could identify 800,000 households likely to buy in a category within 12 months, claiming 94% accuracy without a denominator.

Not the same as

Ad frequency is the average number of times each person in an audience sees an advertisement, and frequency cappinglimits it. It counts exposures, not transactions, though the planning rule of three exposures per purchase cycle, traced to Herbert Krugman in 1972, links the two.

Penetration is the share of households or customers buying at least once in the period. Frequency is measured only among those buyers, so the two multiply rather than overlap.

Repeat purchase rate is the share of customers who bought more than once, a loyalty-stage metric in funnel frameworks. It can rise while average frequency falls, if one-time buyers return once and heavy buyers slow down.

Purchase cycle, or interpurchase time, is the interval between purchases. For regular buyers it is roughly the inverse of frequency, and it governs lookback windows and ad timing.

Recent developments

Purchase frequency is being written into standards and bidding systems. The IAB Europe draft remains open for comment until October 23, 2026, and Google Ads API version 25, released in July 2026, added optimisation for retaining loyalty programme members while removing the older lifecycle goal resources.

Supply is widening beyond retailers, as receipt apps, payment companies and warehouse clubs sell segments built on purchase histories into open-web and retail media buying. The baseline also varies by market: Criteo's June 2026 shopper survey found 69% of recent buyers in Japan and 61% in Germany shopping for groceries daily or several times a week, against 24% in France.

Timeline

  • 1955: Andrew Ehrenberg begins working on consumer panels
  • March 1959: Ehrenberg publishes The Pattern of Consumer Purchases, fitting purchase counts to the negative binomial distribution
  • 1963: William McPhee coins the term double jeopardy in Formal Theories of Mass Behavior
  • 1972: North-Holland publishes Ehrenberg's Repeat-Buying: Theory and Applications
  • October 1984: Goodhardt, Ehrenberg and Chatfield publish the Dirichlet model of purchase incidence and brand choice
  • 1987: Schmittlein, Morrison and Colombo publish the Pareto/NBD model in Management Science
  • 1988: Second edition of Repeat Buying calls purchase frequency per buyer the most basic repeat-buying measure
  • July 1990: Ehrenberg, Goodhardt and Barwise publish Double Jeopardy Revisited in the Journal of Marketing
  • 2005: Fader, Hardie and Lee publish the beta-geometric/NBD model
  • 2010: Oxford University Press publishes Byron Sharp's How Brands Grow
  • July 9, 2012: Charles Graham sets out the w(1-b) double jeopardy relationship for The Marketing Society
  • January 25, 2018: Campaign reports the dispute between Dunnhumby and Sharp over heavy buyers
  • May 14, 2019: Kantar's Brand Footprint shows Colgate at 60.5% penetration and 5.5 purchases per buyer, Indomie at 4.7% and 33.1
  • December 2019: Sharp, Romaniuk and Graham publish Marketing's 60/20 Pareto Law
  • October 15, 2024: Amazon Ads introduces long-term sales metrics for new-to-brand shoppers, defined by a 12-month lookback
  • 2025: Journal of Business Research study of 474 US categories finds purchase frequency changes had minimal impact on category revenue
  • January 2025: Numerator combines with Kantar's Worldpanel
  • April 8, 2025: Google announces lifecycle modes including High Value Win-Back
  • July 1, 2025: Numerator launches the Worldpanel by Numerator brand
  • September 2025: PayPal Ads opens transaction-based audiences, including purchase frequency, to PubMatic buyers
  • November 2025: Amazon Marketing Cloud extends its lookback window from 13 to 25 months
  • December 10, 2025: IAB Australia's Retail Media Council publishes its audience targeting blueprint, including RFM
  • April 2026: Media.net launches Fetch receipt-based audiences for high-frequency categories
  • July 2026: Google Ads API version 25 adds loyalty programme retention optimisation
  • August 2026: Amazon discloses that fresh grocery buyers shop roughly twice as often
  • September 17, 2026: IAB Europe releases a draft in-store standard with purchase cycle lookback tiers
  • September 2026: Sam's Club introduces predictive targeting products; Criteo publishes CPG Pulse 2026
  • October 23, 2026: Comment deadline for the IAB Europe draft

Summary

Who: Brand owners, retailers, agencies and platforms measure purchase frequency; consumer panels such as Worldpanel by Numerator publish it; retail media networks, payment companies such as PayPal and receipt apps such as Fetch sell segments built on it; and researchers at the Ehrenberg-Bass Institute have produced much of the evidence on its limits.

What: The average number of purchases per buyer of a brand, product or category in a set period, calculated as purchases divided by buyers. Multiplied by penetration, it gives purchase occasions; multiplied by order value and lifespan, it gives customer lifetime value.

When: Ehrenberg modelled purchase counts in 1959, the Dirichlet linked frequency to brand size in 1984, and customer-level models followed in 1987 and 2005. Between 2024 and 2026, retail media networks, bidding systems and draft standards adopted it as a targeting, bidding and lookback input.

Where: In household panel data, e-commerce and analytics platforms such as Shopify and Google Analytics 4, retail media networks including Amazon Ads and Sam's Club, programmatic curation through Media.net and PubMatic, and IAB Europe's draft in-store measurement standard.

Why: Sales growth can come from more buyers or from buyers returning more often, and the split determines where budgets go. Evidence from panel research suggests frequency moves far less than penetration, while retailers and data sellers market frequency as the lever they can deliver, so the metric sits at the centre of a live dispute over how brands grow.