What is RFM segmentation?

RFM segmentation sorts customers by three dimensions of behaviour: how recent the qualifying event was, how often it happens, and how much they spend.

That much every published guide agrees on. Almost nothing after it is settled.

RFM segmentation, as published on 2026-09-03. One platform scores each dimension 1 to 3. Most score 1 to 5. One recommends four tiers and advises against going beyond four. One reference model uses 1 to 10. In most systems 555 is the best customer and 111 the one you stop marketing to. In one, 1-1-1 is the shorthand for Best Customers. Monetary value means total spend in five sources, average order value in two, and the single largest order in one. And under the most common scoring method, a fifth of your customers score 1 no matter how well the business is doing.

What is RFM segmentation?

It is a way of ranking customers using their behavioural and transaction history.

Each customer gets a score on three dimensions. The scores are combined, and customers with similar combinations are grouped and marketed to differently. Wikipedia places it in database and direct marketing. Optimove gives the reason it stuck: the scales are numeric, and no data scientist is needed to produce it.

That simplicity is real. It is also why so many incompatible versions exist.

What do recency, frequency and monetary actually measure?

Recency is the closest to settled, and even it is not.

Every source that defines it measures elapsed time, almost always in days. What they disagree about is which event stops the clock. Three of them define it as the last purchase. Braze counts the last time someone engaged or bought, and Optimove counts the last activity of any kind — usually a purchase, sometimes a site visit or an app session.

Frequency and monetary value are further apart. Frequency is a count of activity, usually orders, but not always; some sources count interactions too. The window is another setting nobody agrees on. Monetary value is spend, but sources disagree about which spend they mean. Both are covered below, because the difference changes who lands in which group.

Is there a standard RFM score?

No. Four different scales are in published use.

Klaviyo scores each dimension 1 to 3. Shopify, Braze, Rejoiner and Yotpo all use 1 to 5. Optimove recommends four tiers and says going beyond four is not worth the added difficulty. Wikipedia’s worked example runs 1 to 10. Mailchimp declines to fix it at all: most companies use 1 to 5, but you can use any values you find useful.

So a “score of 3” is a top mark in one system, the exact middle in another, and below the midpoint in a third.

Does a higher score always mean a better customer?

Usually, but not always, and the exception is not marked.

In most published models the highest number is the best behaviour. The most recent buyer takes the top score, and so do the most frequent buyer and the biggest spender.

One model runs the other way. In the four-tier scheme above, tier 1 is the most recent, the most frequent and the highest spend, and tier 4 is the least. Its best customers sit in tier 1 on all three dimensions.

Nothing in the notation tells you which convention you are looking at. The digits look identical either way.

What does an RFM score of 111 mean?

It depends entirely on who calculated it.

Rejoiner is blunt about its own version: the customer ranked 555 is your most recent, most frequent and biggest spender, and 111 is the opposite, “probably not worth marketing to”. Yotpo and Braze say the same. On a 1-3 scale, Klaviyo’s 111 group is labelled Inactive.

But in Optimove’s tier notation, 1-1-1 is the shorthand for Best Customers.

Same three characters. Opposite customer. If a report, an agency or a dashboard hands you an RFM code without naming its scale and its direction, the code does not tell you anything yet.

How are the three scores combined into one?

Four different ways, producing four different outputs.

Five sources join the three scores into one string, so a top-of-scale customer on a 1-to-5 model is written 555. Some add the scores instead, giving a total between 3 and 15. Some average them, giving a figure between 1 and 5.

Shopify’s own reporting does none of those. It keeps recency on its own, then averages only frequency and monetary value as (F + M) / 2, and rounds the result down. A customer with frequency 2 and monetary 3 gets an FM of 2.5, which becomes 2.

That rounding matters. Two customers who are mirror images — one who orders often and spends little, one who orders rarely and spends a lot — collapse to the same value, and ties break downward.

Does monetary value mean total spend or average order value?

Three answers are in print. Two vendors publish two of them themselves.

Most of the sources that define it mean lifetime or period total spend. Rejoiner sums revenue across the customer’s lifetime. The four-tier guide sums it across a defined period, and treats spend-per-order as a separate secondary factor rather than the metric itself.

Omnisend means something different: monetary is the customer’s average order value. Shopify’s guide agrees, defining it as how much customers spend per order — while the help centre for its own product, linked above, defines it as the total amount spent. Wikipedia’s example uses neither, scoring the single highest purchase.

The gap is not small. Ten $50 orders and one $500 order are identical on total spend, ten to one apart on average order value, and ranked opposite ways on largest order.

Over what time window is frequency counted?

Anywhere from twelve months to forever.

Omnisend counts only orders from the last 365 days, and says older orders stop affecting the result. Klaviyo counts within the report’s own time frame. Rejoiner counts the customer’s whole lifetime. Braze and Optimove leave the window to the operator, which makes it a setting rather than a definition.

Take a customer with ten orders in 2024 and none since. Their frequency is 10 in a lifetime model and 0 in a trailing-year one. Both numbers are correct. They describe the same person.

This is the same lifetime-versus-window split that divides the published definitions of repeat purchase rate, and it has the same effect: the metric name travels between businesses and the number underneath it does not.

Are RFM scores relative to your own file or absolute?

Mostly relative, which is the part that surprises people.

The common method is quintile scoring. You sort your customers on a dimension, cut the list into five equal groups, and the top fifth scores 5. Braze, Rejoiner and Bloomreach all work this way. Shopify states the consequence plainly: a 5 means the top 20% of that dimension for your store, and the scores are based only on your store’s data, not on industry standards.

Klaviyo mixes the two. Its frequency and monetary scores are percentile-based — top, middle and bottom third. Its recency score is not. Recency is a fixed calendar rule: within 180 days scores 3, within 365 days scores 2, and beyond 365 days scores 1.

Why does someone always score a 1?

Because quintile scoring creates the bottom group before it looks at anyone.

If a dimension is cut into five equal parts, exactly 20% of the file lands in the lowest part. That is arithmetic, not a finding about customers. A store where every single customer bought last week still has a fifth of them scoring 1 on recency.

So a low quintile score is a position in a queue. It is not evidence that a customer has lapsed. Under a fixed-threshold rule the same digit means something real — no purchase in over a year is a fact about the person, and it stays true however healthy the rest of the file is.

Reading one as the other is an expensive mistake. It routes win-back offers to people who are not lapsed at all.

Can you compare RFM scores between two stores?

Only if both used absolute thresholds, and most do not.

A relative score describes where someone sits among your customers. Move the same person to a different store with a different customer base and their score changes without their behaviour changing at all.

This makes RFM scores a poor benchmark between businesses. A claim about the share of your file scoring in the top bands tells another operator nothing, because quintile scoring fixes that share in advance. The number describes the method.

How many RFM segments are there?

Published counts run from 6 to 125.

Klaviyo defines six groups and fixes their names. Shopify defines eleven, one of which is Prospects — customers with no orders at all, and therefore no recency, frequency or monetary value to score. The four-tier model produces 64 combinations, or 27 if three tiers are used. Bloomreach’s use-case page says eleven segments twice, then lists twelve names; its product documentation carries a numbered table of twelve.

Why do the segment names not match the definitions?

Because the vocabulary converged and the underlying rules did not.

Segment names repeat across vendors without the rules underneath them matching. “At risk” is defined five different ways across five sources: as a specific set of score codes, as a recency-and-FM range, as a long list of three-digit combinations, as a purchase 90 to 180 days ago, and as a customer with strong history who is past their normal buying gap.

Two published models also reach outside RFM to finish the job. Bloomreach’s bottom two segments carry an identical score set, and are separated only by whether the customer clicked an email or visited the site recently. Three of Rejoiner’s twelve cohorts are defined on browsing and email behaviour rather than on purchases.

So a shared segment name is not a shared segment.

What do you need before you can run an RFM analysis?

Far more in a platform than on paper.

The method itself needs a customer list, order dates and order values. One guide says a .csv export is enough to start, and another says a spreadsheet will do.

Platform implementations set real floors. Klaviyo’s report requires at least 500 customers who have placed an order, 180 days of order history, orders in the last 30 days, and some customers with three or more orders — and it sits behind a paid add-on rather than the core product. Omnisend says its map is most accurate above roughly 100 returning customers, and falls back to default day thresholds below that.

A young store can do the arithmetic and still not qualify for the report.

How often should RFM scores be recalculated?

Published cadences run from hourly to monthly.

Omnisend recalculates stages once an hour. Klaviyo’s academy guide says profile data updates daily, while its segment properties are grouped by month. Bloomreach’s own documentation gives three different answers on one page: twice per month, every month, and monthly.

Shopify’s guidance is the one tied to the business rather than the platform: update at least as often as your typical purchase cycle. Monthly for a monthly cycle, quarterly for a slower one.

Where did RFM come from?

Two published origin stories, about sixty years apart.

Rejoiner dates it to catalogue direct mail in the 1930s and 1940s, when merchants kept a 3x5 index card per customer and ranked the file by hand before paying to print and post. Investopedia dates it to the 1990s, and points to a 1995 Marketing Science paper by Bult and Wansbeek, Optimal Selection for Direct Mail.

Both agree the setting was direct mail. Neither cites the other’s evidence, so the honest reading is that the date is unsettled and the origin is not.

What can RFM not tell you?

Every source that discusses limits names the same one.

It uses three variables, and all of them are historical. Mailchimp puts it directly: because RFM uses historical data, it may not predict future behaviour.

Wikipedia adds a warning the vendor pages do not: RFM segments can fit chance relationships in the data they were built from, so the result should be validated against a separate set of data before it is trusted.

One operational catch is worth knowing. Klaviyo notes that RFM group membership ignores marketing consent entirely, so a group is not a sendable audience until you filter it.

How should you read an RFM score you did not calculate yourself?

Ask six questions before you act on it.

1. What is the scale, and which end is good? Without both, a code like 111 is unreadable.
2. What event counts? Some models count purchases only. Others count engagement, sessions or app use too.
3. How are the scores combined? A concatenated code, a sum, an average and an R-by-FM grid are different outputs.
4. Is the score relative or absolute? A quintile 1 is a rank. A threshold-based 1 is a fact.
5. What does monetary mean here? Total spend, average order value and largest order rank your customers differently.
6. What window does frequency use? Lifetime and trailing-year models disagree about the same person.

None of this makes RFM a weak method. It is cheap, it runs on data you already have, and it turns a flat customer list into something you can act on. But the score is only meaningful inside the model that produced it.

If you are setting one up, the useful work is not choosing a tool. It is writing down your own answers to those six questions first, so that the groups mean the same thing in six months as they do on the day you build them.