Most businesses know their revenue by product line and by month. Few know it by type of customer, and that is where the money is decided. RFM is the simplest way to see it: three numbers per customer, computed from a plain transaction log. It works the same for a fund’s players or backers, the clients of an accounting firm, a clinic’s patients or a subscription base. On a public example dataset, 33.4% of customers bring 75.0% of revenue.
The data
To keep every number checkable, we use a public example dataset: UCI Online Retail II, the transaction log of a UK online retailer whose buyers are mostly small businesses. We use 2010-12-01 to 2011-12-09: 541,910 rows, customers in 38 countries. The method needs only three fields per transaction: who, when and how much.
We removed cancellations, zero or negative lines, postage, manual adjustments, duplicates and lines without a customer ID. That leaves 4,334 customers and $8,737,228 of revenue.
How RFM scoring works
Each customer gets three numbers, measured on 2011-12-10, the day after the last transaction:
- Recency: days since the last purchase.
- Frequency: number of separate purchases.
- Monetary: total spend.
Each number is split into quintiles and scored 1 to 5, where 5 is best. The recency × frequency grid places every customer in one of ten standard segments: Champions (recent and frequent), Loyal, At risk and Can’t lose (frequent customers gone quiet), newcomers such as New and Promising, and Hibernating (one or two purchases, long ago).
Where the revenue sits
Source: UCI Online Retail II (Chen, 2019), sheet Year 2010-2011, CC BY 4.0; our calculation
Champions are 14.6% of customers and bring 49% of revenue; with Loyal customers that is 33.4% and 75.0%. The top 20% of customers by spend bring 74.6%. Hibernating is the largest group, 24.9% of the base, with 5.9% of revenue.
| Segment | % of customers | % of revenue | Median purchases | Median days since last purchase |
|---|---|---|---|---|
| Champions | 14.6% | 49.0% | 8 | 5 |
| Loyal | 18.8% | 26.0% | 5 | 30 |
| Can't lose | 1.5% | 2.0% | 7 | 106 |
| At risk | 13.7% | 7.3% | 3 | 135 |
| Potential loyalists | 11.1% | 5.6% | 2 | 18 |
| Need attention | 4.4% | 1.8% | 2 | 52 |
| New | 1.0% | 0.2% | 1 | 8 |
| Promising | 2.2% | 0.3% | 1 | 23 |
| About to sleep | 8.0% | 1.9% | 1 | 53 |
| Hibernating | 24.9% | 5.9% | 1 | 219 |
Put more simply: customers with two or more purchases are 65.3% of the base and bring 92.8% of revenue. Here the second purchase matters more than the next new customer, and in most repeat-customer businesses the pattern is similar.
Do customers come back?
A cohort is everyone whose first purchase fell in the same month; retention in month N is the share of it that bought again N months later. December 2011 is excluded because it is only partly recorded.
- Dec 2010 cohort
- All cohorts, average
Source: UCI Online Retail II (Chen, 2019), sheet Year 2010-2011, CC BY 4.0; our calculation
The average cohort keeps about a quarter of its customers: 23.5% buy again in month 1 and 28.8% in month 6. Retention does not decay toward zero.
The December 2010 cohort looks much stronger, 36.5% in month 1 and 50.2% in month 11. That is history and season, not loyalty: the data starts that month, so its “new” customers include long-standing ones, and month 11 is the seasonal peak. A cohort chart that ignores both would credit a campaign that never ran.
What to do with each segment
- Champions, Loyal: protect. A named contact, priority service, no blanket discounts.
- Can’t lose, At risk: win back. Personal outreach within weeks. The 65 “Can’t lose” customers had a median of 7 purchases before going silent.
- Potential loyalists, New, Promising: get the second visit. A timely reminder and a clear next step.
- Need attention, About to sleep: re-engage. Test one offer against a control group first.
- Hibernating: spend little. Low-cost reactivation at most.
Limits
- One business, one year, customers without an ID excluded; a business with irregular or seasonal demand will look less regular.
- Quintiles are relative: a “Champion” is a top customer of this business, not a fixed spend threshold.
- RFM describes, it does not predict. For who will come back next you need a model, like the retention model in our case study.
Download the template
RFM segmentation template (Excel, .xlsx): no sign-up, no form. Paste your orders (customer ID, date, order ID, value), list each customer once, and the formulas fill in recency, frequency and monetary value, the 1–5 scores, the segment and a revenue summary by segment with what to do about each one. It opens in Excel, LibreOffice and Google Sheets, has no macros and comes with a sample from the same public dataset so you can see it working before you replace the data.
The template scores by the share of customers with a worse value, so tied customers share a score; the script behind this article breaks ties by rank. On your own data the two can place a few borderline customers in neighbouring segments. Per-customer RFM features for the whole dataset are in the customer retention dataset.
Data: UCI Online Retail II (Chen, 2019), licensed CC BY 4.0. All numbers on this page are calculated by our script on that dataset and loaded into the page from its output.