Blog
RFM Analysis: Which Customers Should You Contact Next?
Two customers have each spent €600 with you. They sit next to each other in your sales report, so you send them the same offer.
One bought a big training package three days ago. She could use a hand booking her first session. The other used to train with you every week. His last visit was six weeks ago. What would you say to him? Probably something a little more personal than “Ready to get started?”
RFM analysis groups customers by when they last bought, how often they buy, and how much they spend. You can try it with a booking calendar or an order history. The point is to walk away with someone's name and a good reason to get in touch.
What does RFM stand for?
Measure | Question | What to record |
|---|---|---|
R — Recency | How long ago did this customer last buy? | Days from the last purchase to the date of your analysis |
F — Frequency | How many times did they buy during the period? | Number of distinct purchases |
M — Monetary value | How much did they spend during that period? | Total purchase value |
Those are the three dimensions used in Adobe Commerce's RFM example. You choose a time window and a scoring system that fit your business.
Suppose someone hasn't ordered coffee in two months. Are they disappointed? On vacation? Working their way through the enormous bag of beans you sold them? Their order history tells you when to start asking. The answer takes a little more digging.
Step 1. Decide what counts as a repeat purchase
Three items in one shopping cart make one order. A ten-session training package is one purchase, even if the client shows up ten times. Attendance is worth recording separately. It tells you whether the client is using what they bought.
The same logic applies to a creator who shoots product videos for brands. Five videos commissioned for one campaign make one order. A deposit and a final payment on that campaign still make one order. If you want to know whether clients come back, count the decisions they make to hire you again.
Compare people who've had similar opportunities to buy. Give a coffee machine buyer and a monthly coffee bean subscriber the same “60 days without an order” flag and you'll spend a lot of time worrying about perfectly happy espresso machine owners.
Step 2. Clean up the order history
Pick an analysis date and a period for counting purchases and spending. Ninety days might cover several chances to buy another training session. For a creator working on seasonal campaigns, a year may make more sense.
Start with four fields per order — customer ID, order ID, purchase date, and amount. Then:
- Match orders made by the same person or business under different names.
- Remove duplicates and cancellations. Count each order once, however many items or payments it contains.
- Use the amount after discounts and refunds. Keep everything in one currency and handle tax consistently.
- Decide what to do with fully refunded orders. For this exercise you can exclude them from recency and frequency and give them zero net sales. Keep the costs they incurred for your profitability analysis. A partial refund reduces spending, while the order still counts once. Revisit recent orders when their return window closes.
Measure recency against the same date for everyone. Use the same period for both frequency and spending. Someone who bought 100 days ago can have zero purchases and zero spending in a 90-day window, with a recency of 100 days. If the date is unknown, leave it missing. Zero days means they bought today.
Add the date of the first-ever purchase if you have it. You'll need it to recognize newcomers. And pay attention to whose records you have. A café that can identify only loyalty card holders has learned something about its card holders. The people in the rest of the line are still a mystery.
Step 3. Calculate RFM for five training clients
Take 90 days of bookings for a personal trainer. Everyone in this example pays per session, at €50 a visit, with no discounts or refunds. I made up the price. Your trainer is welcome to charge more.
Client | R: days since last purchase | F: purchases in 90 days | M: spending in 90 days |
|---|---|---|---|
Sara | 7 | 16 | €800 |
Tomek | 42 | 12 | €600 |
Ola | 3 | 1 | €50 |
Ewa | 10 | 4 | €200 |
Jan | 60 | 2 | €100 |
Tomek spent €600 and hasn't been back in six weeks. Ola spent €50 and only just started. Sara bought 16 sessions and was here last week. If you have a dozen clients, you may already know whom to call. It's fine to stop here.
Step 4. Give each measure a score
With hundreds of clients, you'll want a quicker way to spot patterns. Start with three scores. A 3 means a recent purchase, frequent purchases, or high spending. Watch the direction of the first column. Fewer days since the last purchase earn a higher recency score.
Suppose this trainer's active clients normally return every week or two. Here's one possible set of cutoffs.
Score | R: days since last purchase | F: purchases in 90 days | M: spending in 90 days |
|---|---|---|---|
3 | 0–14 | 12 or more | €600 or more |
2 | 15–28 | 4–11 | €200 to under €600 |
1 | 29 or more | 0–3 | Under €200 |
Check the gaps between your own customers' purchases before borrowing these numbers. A month of silence can be perfectly ordinary between video campaigns. It means something else at a bakery where you usually buy breakfast every morning.
Client | R score | F score | M score | RFM profile |
|---|---|---|---|---|
Sara | 3 | 3 | 3 | 333 |
Tomek | 1 | 3 | 3 | 133 |
Ola | 3 | 1 | 1 | 311 |
Ewa | 3 | 2 | 2 | 322 |
Jan | 1 | 1 | 1 | 111 |
Read 133 one digit at a time. Tomek last bought a while ago, bought frequently during the period, and spent a lot. IBM's scoring example uses the same convention of putting the three digits side by side. IBM SPSS documentation
Adding the digits would lose something important. A 311 and a 113 both sum to five. The first might be someone who made a small purchase yesterday. The second might be someone who bought one expensive service a long time ago. Would you send them the same email?
In our trainer's example, every session costs €50, so spending tells you exactly what frequency already told you. Once some clients buy premium sessions or extras, the two columns can tell different stories.
What about scoring from 1 to 5?
You can rank customers separately on each measure and split each ranking into five groups, as in Adobe's example above. That tells you where someone stands relative to everyone else. Be careful with ties. If nearly everyone has bought once, distributing them across five frequency bands can manufacture distinctions that mean nothing. For a small business, start with cutoffs you could explain to the person who runs your calendar.
Step 5. Turn the profiles into follow-ups
You don't need a campaign for every possible code. First set aside incomplete records for a closer look. Then pull out newcomers. For this trainer, we'll count someone as new if their first-ever purchase was in the past 14 days and it's the only purchase on record. That's Ola.
Use the rules below in order. Here R and F refer to scores.
Group | Rule | First action |
|---|---|---|
New | First-ever purchase in the past 14 days, only one purchase to date | Ask how the first session went and agree on a next step. |
Former regulars who've gone quiet | After removing newcomers, R = 1 and F = 2 or 3 | Find out what interrupted the routine. This is Tomek. |
A gap worth checking | R = 2, any F | Look for an upcoming booking or a known reason for the gap. |
Active repeat customers | R = 3 and F = 2 or 3 | Make it easy for Sara and Ewa to keep a regular slot. |
Other occasional or inactive customers | Of those remaining, R = 1 or 3 and F = 1 | Check whether a low-cost follow-up makes sense. |
Spending can help you decide whom to look at first within each group. An old client who spent a lot might warrant a call. To decide how much you can afford to offer, you'll also need your margin and service costs.
The trainer now has a short list. Ask Tomek what happened. Help Ola book her second session. Keep Sara's usual time free. Before sending Jan an offer, check what he hired you for. He may have achieved exactly what he wanted.
How RFM changes the conversation in different businesses
At a hair salon, compare clients who book services with similar intervals. If a regular is overdue, look at next week's calendar. The fastest way to spoil a thoughtful “We miss you” message is to send it to someone who's already booked for Friday. Offer an easy way to book, then check how many came back and what those visits earned after the cost of the offer. A discount makes sense when you have reason to think price is the obstacle.
For a freelance video creator, count separate campaigns over the past year. A regular brand client has gone unusually quiet. Ask what launches are on the calendar before spending a day preparing a pitch. The last project may simply have ended. For retainers or monthly subscriptions, payment schedules make everyone's R and F look similar. Renewal dates, actual use, and service hours may tell you much more.
In an online cosmetics store, a first-time buyer may appreciate instructions for getting the best out of a product. A repeat buyer approaching the bottom of the jar may appreciate a link to their usual order. Give a three-month supply time to run out. And if your main product gets replaced once every five years, look for service needs, accessories, or referrals between purchases.
Does a high RFM score mean a profitable customer?
A customer can score 333 while buying only on promotion, returning merchandise, or using hours of staff time. M measures customer spending. To calculate what the relationship earns you, you also need the cost of fulfilling and supporting those orders.
I use the customer loyalty matrix alongside RFM for exactly that reason. RFM can tell you when it might be worth getting in touch. The matrix helps you decide how much time and money to put into the relationship. A 333 alone won't settle that question.
Try it on one buying cycle
Pick one kind of offer. Look at a few dozen customer records and write down the analysis date, window, purchase unit, and three measures. Find a handful of people who used to buy regularly and have passed their usual return date. Check their circumstances and try one follow-up. Record what it cost and when you'd expect a return visit.
With a large enough group, randomly select some comparable customers to keep receiving ordinary service without the new offer. After the same amount of time, compare return rates and contribution per customer in both groups, including the cost of the follow-up. Some people would have returned anyway. With a handful of customers, treat the result as a clue for the next round.
After the next buying cycle, look at the raw numbers again. Who came back? What did those orders earn after costs? How much work did they take? A shift in scores tells you little if you also moved the cutoffs.
Start with the regular whose name you used to see in your calendar every week. RFM can find the empty slot. A conversation can tell you what happened.
Newsletter
Want more essays like this one?
One useful idea at a time — on brands, choices, stories and AI. Written when there is something worth developing, not to fill a schedule.

