RFM analysis is a way of scoring every customer on three things: how recently they bought, how often they buy, and how much they have spent.
If you sell anything people buy more than once, those three numbers are already sitting in your store or CRM. RFM turns them into a ranked list, so you can send a first-time buyer something different from what you send your best customer.
Most brands segment on the last product someone ordered instead. You pick a name like "engaged subscribers", draw the line wherever feels about right, and every campaign after that rests on a definition nobody wrote down.
That guesswork costs money. You discount people who were going to buy anyway, you send win-back emails to customers who ordered last week, and your best customers learn that waiting produces a coupon.
The good news is that RFM takes an afternoon in a spreadsheet. It works whether you have 800 customers or 800,000, and once it is set up you can rerun it every month.
This guide covers what the three scores measure, how to build them without a data team, the segments they produce, what to send each one, how to tell whether it worked, and where the model runs out of road.
tl;dr
- Product tags describe what someone bought. RFM scores describe whether the relationship is still alive.
- Score Recency, Frequency and Monetary from 1 to 5, then collapse the combinations into eight or so working segments.
- Champions respond to access. At Risk customers respond to a reason. Lost profiles belong off your campaign sends entirely.
- Build it from purchase properties you already have, then measure at 28 and 56 days against a list-wide control.
- Recency doubles as a deliverability signal, so quiet profiles cost you inbox placement as well as revenue.
What RFM Measures
Recency is days since the last order. This carries the most weight of the three. Someone who ordered eleven days ago is in a live conversation with your brand. Someone who ordered 187 days ago is a different person to write to.
Frequency is order count inside your scoring window, usually twelve months. One order is a trial. Four orders is a habit, and the jump from one to two is the largest behavioural step most stores ever see.
Monetary is total spend across that same window. Pick total spend or average order value and stay consistent, because you will rerun this every month. Total spend surfaces whales. Average order value surfaces people who buy richly but rarely.
Score each dimension from 1 to 5, where 5 is best.
| Score | Recency | Frequency (12 mo) | Monetary (12 mo) |
|---|---|---|---|
| 5 | 0 to 30 days | 5+ orders | Top 20% |
| 4 | 31 to 60 days | 3 to 4 orders | Next 20% |
| 3 | 61 to 120 days | 2 orders | Middle 20% |
| 2 | 121 to 180 days | 1 order | Next 20% |
| 1 | 181+ days | none in window | Bottom 20% |
Treat those day buckets as a starting point. A weekly coffee subscription and a mattress brand need different Recency scales. Set yours from your own median days to second order and move on.
RFM is behavioural segmentation reduced to the three behaviours that predict revenue best. Age, location and job title stay out of it deliberately.
Why Product Tags Fall Short as a Campaign Brain
"Bought the serum" is useful inside a post-purchase flow. It is thin as a way of deciding who receives Tuesday's campaign.
Picture two people who bought the same serum. The first ordered nine days ago, has six orders this year and has spent $640. The second ordered once, 140 days ago, and spent $42. The first wants early access to the next drop. The second needs a reason to place a second order at all.
Product data decides which block renders inside the email. RFM decides who receives it in the first place.
Revenue per recipient is the number worth watching here. Open rate tells you a subject line worked. Revenue per recipient tells you the audience was right.
How to Score Your List Without a Data Team
Three fields and a mapping table will do it.
Step 1: export four columns. Customer email, last order date, order count and total revenue, over the last twelve months unless your purchase cycle is shorter. Every major platform exports this.
Step 2: sort into fifths. Sort by recency, cut into five equal groups, and give the most recent fifth a 5 and the oldest fifth a 1. Repeat for frequency, then for monetary. Every customer now carries three digits.
Quintiles beat round-number thresholds once your list is large enough, because they stay honest when your business changes. If average spend drops next quarter, a 5 still means top 20% of your customers. Below roughly 2,000 buyers, use fixed buckets like the table above so that one large order does not warp the whole scale.
Step 3: read the digits as a pattern, not a total. Adding them together destroys the signal. A 5-1-1 bought once, recently, and spent little. A 1-5-5 bought constantly, spent heavily, and has gone silent for a year. Both average to something middling, and they need opposite emails. Read left to right, starting with recency, because recency tells you whether the relationship is still alive.
Step 4: rerun monthly. Recency scores drop every day someone does not buy, so a segment built in January describes people who have moved on by April.
The Segments Your Scores Hand You
Five scores across three dimensions produce 125 combinations, which is how teams end up with a spreadsheet and no campaign. Collapse them into eight working groups.
| Segment | Typical pattern | What they need |
|---|---|---|
| Champions | R4-5, F4-5, M4-5 | Access, early drops, loyalty perks |
| Loyal | R3-5, F4-5, M3-5 | Habit protection, replenishment timing |
| Potential loyalists | R4-5, F2-3, M2-4 | A frictionless second and third order |
| New customers | R5, F1 | Onboarding that earns order two |
| Need attention | R3, F3-5, M3-5 | A specific question, not a blast |
| At risk | R1-2, F3-5, M3-5 | A named reason to return |
| Hibernating | R1-2, F1-2, M1-3 | One short sunset sequence |
| Lost | R1, F1, M1 | Removal from campaign sends |
The pattern in that chart shows up on most audits. A small group of champions and loyal customers carries the majority of email revenue, while the biggest block of profiles by headcount contributes almost nothing and still receives every send.
What to Send Each Segment
Champions get access. Early releases, restocks before the public announcement, loyalty perks and referral asks. They already buy at full price, so a sitewide discount code hands away margin you had earned. Put the code in a conditional block and hide it from anyone scoring 4 or 5 on recency and monetary.
Loyal customers get relevance. Time the email to their replenishment cycle rather than your campaign calendar, and show the adjacent product that people like them buy on order three or four. This is where personalisation repays the setup effort, because you finally have enough history to be accurate.
New customers and potential loyalists get onboarding. Three to five emails over ten to fourteen days that deliver on the original promise, show one useful outcome, prove other people got it, then ask for the second order. The structure in email onboarding best practices applies directly, and a well-built post-purchase sequence does more for this group than any discount.
Need attention gets a question. One product or one direct question, not a twelve-product grid. A subject line like "You usually reorder around now" outperforms a generic promo, and if they click without buying, wait before sending again.
At risk customers get a named reason. Say the gap out loud: "It has been four months since your last order." Then give a reason to return that is not purely price. A restock of the item they bought, a fix you made, a better size guide. Hold the heavy offer for the second or third email, because leading with 25% off teaches every future lapse that disappearing pays. A structured win-back campaign belongs here rather than in your weekly promo, and a re-engagement sequence covers the mechanics.
Hibernating profiles get a sunset. Two or three emails, a preference centre, then removal from campaigns. This is list cleaning with a final conversation attached, and your sunset policy should define the exit.
Lost profiles come off campaign sends. Keep transactional email running and keep a capped win-back flow if you want one. Sending weekly promos to profiles that reliably ignore you buys cheap opens and expensive reputation damage.
Building It in Klaviyo
Everything you need already exists on the profile.
Build recency from Last Purchased, frequency from Placed Order count over your window, and monetary from Historic Customer Lifetime Value or the sum of placed orders. Klaviyo also ships an RFM report, so screenshot the current distribution before you change anything. That screenshot is your baseline. If you would rather have the scoring and the campaign gating built for you, that is standard Klaviyo agency work.
Then gate the campaigns:
- Default campaign audience becomes champions, loyal, potential loyalists and need attention
- New customers and promising buyers stay in flows for their first fourteen days
- At risk gets its own three-email track with a cooldown, separate from the weekly promo
- Hibernating and lost are excluded from campaign sends
Change who enters the weekly campaign first, before rebuilding a single flow. That one audience swap carries most of the benefit and takes an afternoon.
How to Tell Whether It Worked
Snapshot before you change anything, then measure a curve rather than a feeling.
At day 0, record list-wide sends, revenue, revenue per recipient, unsubscribe rate and spam complaint rate for the previous 28 days. Record the same per segment. Note any promotions in the calendar so a seasonal spike does not get credited to segmentation.
At day 28, pull the same rows and ask three questions. Did champion revenue per recipient hold once you removed the discounts? Did at-risk revenue rise without the rest of the list becoming more discount-dependent? Did complaint rate fall after lost profiles came off campaigns?
At day 56, repeat. Twenty-eight days can be noise. Fifty-six days shows direction.
Always compare against the account as a whole over the same dates. Measuring this month against last month will mislead you the first time a product launch or a seasonal swing moves everything at once. Email marketing KPIs covers which numbers deserve a place on that dashboard.
If sends fall and revenue falls with them, your gate was too tight. Add potential loyalists back first, and leave lost profiles where they are.
Common Mistakes and Real Limitations
Scoring people who have never bought. Profiles with zero orders have no honest monetary score. Use recency and frequency of clicks for never-buyers and leave the M out.
Reusing someone else's recency buckets. If your median time to second order is eleven days, calling 90 days "recent" makes the whole scale meaningless.
Discounting champions because the campaign needs a code. Use a conditional block and hide it from your top scorers.
Building segments and mailing none of them. Eight segments is plenty. Ship the campaign gate this week rather than perfecting the model.
Scoring once and never refreshing. Scores decay. Rebuild monthly at minimum, or use live RFM groups so a quiet champion becomes at risk without a quarterly workshop.
The model has real limits too. RFM only describes buyers, so you still need engagement segments for the majority of your list who have never ordered, which is where a broader segmentation strategy comes in. It is backward looking, so someone who just had a terrible support experience still scores 5-5-5 for another month. Subscription businesses need a different frame entirely, because automatic monthly billing turns recency and frequency into billing artefacts rather than behaviour. And on lists under a few hundred buyers, quintiles are so small that a single refund moves someone two tiers.
What Good Looks Like
Recency is also an inbox placement signal, because mailbox providers watch who stops engaging. Your sending rules should follow the scores.
| Tier | Complaint risk | Campaign rule |
|---|---|---|
| Champions and loyal | Lowest | Full cadence |
| New and potential | Low | Flows first, then campaigns |
| Need attention | Medium | Fewer campaigns, higher relevance |
| At risk | Medium to high | Win-back track only |
| Hibernating and lost | Highest | Suppressed from campaigns |
An "engaged" segment that still contains profiles with no clicks in 90 days is a vanity audience, and sender reputation responds to what happens after the send rather than to the name on the segment.
A healthy setup looks like this. Your weekly campaign goes to four segments instead of "engaged 90 days". New buyers finish onboarding before they see a promo. At risk has its own track with a cooldown. Champion emails from the last month contain no sitewide discount codes. And a monthly screenshot of segment counts sits beside your revenue number, so a shrinking champion count becomes visible early. That last one matters most, because a falling champion count is a product or service problem that no amount of segmentation will fix.
To start this week: screenshot your current campaign audience and the revenue per recipient of your last four campaigns. Exclude anyone with no order in 180 days and no click in 90 days from the next send. Hide the discount block from your champions on that same campaign. Move at risk onto a three-email win-back instead of the weekly promo. Then recheck at day 28 with the account-wide numbers sitting next to the campaign numbers.
Four changes, no new tools, and the work stays where your data already lives. If you want the wider plan that sits around this, the Traffic Profit Plan maps where email fits alongside the rest of your acquisition.
FAQ
What is RFM analysis in email marketing?
RFM analysis scores every buyer from 1 to 5 on how recently they ordered, how often they order and how much they have spent. Those three scores are grouped into a handful of segments that determine who receives which campaign.
How is RFM different from regular email segmentation?
Most segmentation starts from a tag, a product or a demographic. RFM starts from purchase behaviour, which predicts future revenue more reliably. You can still layer product and channel data on top of the scores.
Can I use RFM without an ecommerce store?
Yes, as RF rather than RFM. Use last click for recency and click or session count for frequency. Either drop the monetary score or weight clicks on pricing and checkout pages more heavily than clicks on blog posts.
How often should I rescore my list?
Monthly is the slowest cadence worth running, because recency decays daily. Weekly is better if you have built it yourself, and platforms with live RFM groups update continuously.
Which RFM segment should I email most often?
Champions and loyal customers, who have already demonstrated intent with money. New customers and potential loyalists belong in automated flows until they place a second order.
Does RFM replace my welcome series or abandoned cart flow?
No. Those flows keep running on their own triggers. RFM governs campaign audiences and decides which win-back track someone enters.
Will RFM fix a weak offer?
No. If the product has slipped or the site converts poorly, better segmentation will simply reach the wrong conclusion faster. Score the list, then be honest about the offer.