RFM Analysis and Customer Segmentation

What RFM analysis is

RFM analysis is one of the oldest and most effective ways to segment customers. It describes every customer with three numbers taken from their purchase history:

• Recency (R): how many days ago the customer last bought. Recent buyers are far more likely to respond to the next offer.
• Frequency (F): how many orders the customer has placed. Frequent buyers are more engaged and loyal.
• Monetary value (M): how much the customer has spent in total (or per order). High spenders are worth more.

The method comes from direct-mail marketing, where it was found that these three behaviours predicted response to catalogues better than demographics. It is still used by e-commerce, retail, subscription, hospitality and charity fundraising teams because it needs nothing but a transaction file, it is easy to explain, and its segments map directly onto actions: reward the best, win back those who are slipping away, and stop spending on those who are gone.

How the scores are calculated

1. Aggregate orders by customer. For each customer the tool finds the date of the last order, counts the orders and adds up the amounts. Recency is measured in days from the last order to the analysis date, which defaults to the day after the latest order in the data.
2. Score each measure from 1 to 5. Customers are ranked on each measure and divided into quintiles (or quartiles if you prefer a 1–4 scale). For recency, the most recent customers get 5; for frequency and monetary value, the highest values get 5. Customers with tied values receive the same score, which matters for frequency, where many customers have bought only once or twice.
3. Combine into an RFM code. Each customer gets a three-digit code such as 555 (bought very recently, very often, and spent the most) or 115 (spent a lot but long ago and rarely).
4. Assign segments. With 125 possible codes, the tool groups customers into eight segments using the recency score and the average of the frequency and monetary scores, a widely used segment grid.

The eight segments and what to do with them

• Champions (R 4–5, FM 4–5): bought recently, often and spend the most. Reward them, ask for reviews and referrals, give early access to new products. Do not waste discounts on them.
• Loyal customers (R 3–5, FM 3–5): regular, valuable buyers. Upsell, cross-sell and invite them into loyalty programmes.
• New customers (R 4–5, FM 1): first purchase was recent. Onboard them well and give a reason for a second purchase; the second order is the biggest step towards loyalty.
• Potential loyalists (R 3–5, FM 1–2): recent customers with modest frequency or spend. Encourage repeat purchases with bundles, memberships and personalised recommendations.
• Can’t lose them (R 1, FM 4–5): used to be among the best customers but have not bought for a long time. They deserve personal, high-value win-back efforts.
• At risk (R 1–2, FM 3–5): valuable customers who are drifting away. Run reactivation campaigns before they become lost.
• Hibernating (R 2, FM 1–2): low-value customers who have gone quiet. Send low-cost reminders.
• Lost (R 1, FM 1–2): bought once or twice long ago. Include them only in cheap, broad campaigns.

Worked example with the built-in data

The example file contains 3,775 orders from 520 customers over two years. Champions are 28% of customers but generate 56% of revenue, a typical Pareto pattern. Customers in the At risk and Can’t lose them segments together account for 19% of past revenue: they used to buy often or spend a lot but have not ordered for over a year on average, so they are the obvious targets for a win-back campaign. Only 1% are new customers, which tells the business that acquisition has slowed in the last months of the data. The recency × frequency heat map shows the same picture: a bright top-right corner of frequent, recent buyers and a large group of one-time buyers in the bottom-left.

Preparing the transaction file

• One row per order, with a customer ID, the order date and the order amount. If your file has one row per order line, aggregate lines into orders first, or accept that frequency will count lines.
• Dates can be in ISO format (2025-03-14), day-first or month-first formats (14/03/2025, 03/14/2025; the tool detects day-first dates when a day is above 12), or Excel serial numbers.
• Remove cancelled orders and refunds, or net them out, and exclude test accounts and staff purchases.
• Use a long enough window, usually 12 to 24 months, so that frequency means something.

Choosing the options

Analysis date: leave it blank to measure recency from the end of the data, or set it to today’s date if the extract is a few days old. Score groups: quintiles give finer segmentation; quartiles are steadier for small customer bases (a few hundred customers). Monetary value: total spend rewards customers who buy often; average order value separates big-basket shoppers from frequent small buyers, which is useful when frequency is already captured by F.

Using the results

Download the customer scores and import them into your email or CRM platform to target each segment with its suggested action. Track the segment sizes every month: a growing share of champions and loyal customers means retention is working, while a growing at-risk group is an early warning. RFM is also a good starting point for further analysis: use cohort retention analysis to see when customers drop off, and the customer lifetime value tool to predict the future value of each customer rather than describe the past.

Limitations

• RFM is descriptive: it summarises past behaviour and does not predict future purchases. The BG/NBD lifetime value model does that.
• Scores are relative to your own customer base, so a 5 in one business can mean something very different in another, and scores shift as the base changes.
• Equal weighting of R, F and M is a convention. Some businesses weight recency more heavily, since it is usually the strongest predictor of response.
• For businesses with very infrequent purchases (cars, furniture), frequency has little variation; consider recency and monetary value only.

Reporting

Report the period covered, the analysis date, how scores were formed (quintiles of ranks), and for each segment its size, share of revenue and average recency, frequency and spend. Example: “RFM segmentation of 520 customers (3,775 orders, January 2023 to December 2024) identified 145 Champions (27.9% of customers, 55.8% of revenue) and 103 At-risk or Can’t-lose customers accounting for 18.8% of past revenue.”

Key references: Hughes (1994), Strategic Database Marketing; Fader, Hardie and Lee (2005), Journal of Marketing Research, on RFM and customer base analysis; Kumar and Reinartz (2018), Customer Relationship Management.