Most online stores know their revenue by product and month. Few know their revenue by customer type, and that is where the money is decided. We took 541,910 rows of real transactions from a UK online retailer, scored every customer with RFM and found that 33.4% of customers bring 75.0% of revenue.
The data
The dataset is UCI Online Retail II, a public record of a UK gift retailer that sells mostly to small shops. We use the sheet from 2010-12-01 to 2011-12-09: 541,910 rows, customers in 38 countries.
Before scoring we removed what is not a sale: cancelled invoices, lines with zero or negative price or quantity, postage and manual adjustments, and exact duplicate rows. RFM needs a customer ID, so lines without one are left out. That leaves 4,334 customers and $8,737,228 of revenue.
How RFM scoring works
RFM describes each customer with three numbers, measured on the day after the last invoice (2011-12-10):
- Recency: days since the last purchase.
- Frequency: number of separate orders.
- Monetary: total spend.
Each number is split into five equal groups (quintiles) and scored 1 to 5, where 5 is best: most recent, most frequent, highest spend. Recency and frequency place a customer in one of ten standard segments; monetary value is reported for each segment.
| Segment | Recency × frequency score | Who they are |
|---|---|---|
| Champions | R 5, F 4–5 | Bought recently and often |
| Loyal | R 3–4, F 4–5 | Frequent buyers, a little less recent |
| Can't lose | R 1–2, F 5 | Used to buy most often, now silent |
| At risk | R 1–2, F 3–4 | Regulars who stopped buying |
| Potential loyalists | R 4–5, F 2–3 | Recent, buying again |
| Need attention | R 3, F 3 | Average on both, drifting |
| New | R 5, F 1 | First order, very recent |
| Promising | R 4, F 1 | First order, fairly recent |
| About to sleep | R 3, F 1–2 | One or two orders, going quiet |
| Hibernating | R 1–2, F 1–2 | Bought once or twice, long ago |
Where the revenue sits
Source: UCI Online Retail II (Chen, 2019), sheet Year 2010-2011, CC BY 4.0; our calculation
Revenue is concentrated far beyond the usual 80/20 talk. Champions are 14.6% of customers and bring 49% of revenue. Together with Loyal customers that is 33.4% of customers and 75.0% of revenue. The top 20% of customers by spend bring 74.6%.
At the other end, Hibernating customers are the largest group, 24.9% of the base, and bring 5.9% of revenue. They bought once or twice, a median of 219 days before the end of the data.
| Segment | Customers | % of customers | % of revenue | Median orders | Median spend | Median days since last order |
|---|---|---|---|---|---|---|
| Champions | 632 | 14.6% | 49.0% | 8 | $2,599 | 5 |
| Loyal | 814 | 18.8% | 26.0% | 5 | $1,711 | 30 |
| Can't lose | 65 | 1.5% | 2.0% | 7 | $2,006 | 106 |
| At risk | 592 | 13.7% | 7.3% | 3 | $679 | 135 |
| Potential loyalists | 481 | 11.1% | 5.6% | 2 | $511 | 18 |
| Need attention | 190 | 4.4% | 1.8% | 2 | $627 | 52 |
| New | 42 | 1.0% | 0.2% | 1 | $245 | 8 |
| Promising | 94 | 2.2% | 0.3% | 1 | $224 | 23 |
| About to sleep | 347 | 8.0% | 1.9% | 1 | $318 | 53 |
| Hibernating | 1,077 | 24.9% | 5.9% | 1 | $289 | 219 |
Repeat customers carry the business
Split the base more simply, by whether a customer ordered more than once. Customers with two or more orders are 65.3% of the base and bring 92.8% of revenue. The median customer placed 2 orders in the year.
For this store, the question “how do we get more customers?” matters less than “how do we get the second order, and keep the tenth one coming?”
Do customers come back? Cohort retention
A cohort is everyone whose first purchase fell in the same month. Retention in month N is the share of that cohort that bought again N months later. December is excluded from the end of the data because only nine days of December 2011 are recorded.
- Dec 2010 cohort
- All cohorts, average
Source: UCI Online Retail II (Chen, 2019), sheet Year 2010-2011, CC BY 4.0; our calculation
Two things stand out. First, 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; customers who stay, stay.
Second, the December 2010 cohort is much stronger: 36.5% in month 1 and 50.2% in month 11. Part of that is history: the dataset starts in December 2010, so its “new” customers include long-time buyers seen for the first time. Part is season: month 11 is November 2011, the store’s busiest month. A cohort chart that ignores both would credit a retention campaign that never ran.
| First purchase | Customers | Month 1 | Month 3 | Month 6 |
|---|---|---|---|---|
| Dec 2010 | 884 | 36.5% | 38.3% | 36.2% |
| Jan 2011 | 416 | 21.9% | 22.8% | 24.8% |
| Feb 2011 | 380 | 18.7% | 28.7% | 25.5% |
| Mar 2011 | 452 | 14.8% | 19.9% | 26.8% |
| Apr 2011 | 300 | 21.0% | 21.0% | 21.7% |
| May 2011 | 284 | 19.0% | 17.3% | 26.4% |
What to do with each segment
| Segment | Goal | What to try first |
|---|---|---|
| Champions, Loyal | Protect | Early access, account manager, ask what would make them order more. Never send them blanket discounts. |
| Can't lose, At risk | Win back | Personal outreach within weeks, not months. 65 "Can't lose" customers had a median of 7 orders before going silent. |
| Potential loyalists, New, Promising | Get the next order | A reason for the second order soon after the first: replenishment reminders, a related range. |
| Need attention, About to sleep | Re-engage | Test one offer against a control group before rolling it out. |
| Hibernating | Spend little | Low-cost reactivation at most. This is the biggest group and the smallest revenue. |
Limits of this analysis
- One retailer, one year. The store sells mostly wholesale gifts to small shops, so order values are higher and purchases more regular than in a consumer store.
- Customers without an ID are excluded. They account for the gap between total sales and the revenue used here.
- Segments are descriptive. RFM shows who bought what and when; it does not predict who will buy next. For that you need a model, such as the retention model in our case study.
- Quintiles are relative. A “Champion” here is a top customer of this store, not a fixed spend threshold.
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.