RFM analysis

The three attributes

RFM stands for recency, frequency and monetary value. Recency is the gap since the last purchase, frequency the number of purchases in a period, monetary value the revenue or margin in that period. Three numbers produce a score that makes customers comparable without training a model.

The order is not arbitrary. Recency carries the strongest signal, because it predicts best who will respond to an approach. Someone who bought last week is more receptive than someone whose purchase dates back two years, even if the amount was higher then.

How the scoring works

A scale from 1 to 5 per attribute is common, built from fifths of your own customer base. The most active fifth receives a 5 for recency. Three values give a combination such as 5-4-4. From 125 possible combinations, groups are then formed, because nobody works 125 segments.

• Champions: high values in all three attributes. Candidates for referral programmes, not for discounts.
• Loyal: regularly active, medium volume. The core that has to be kept.
• Promising: active but rare and small. Growth comes from basket expansion, not from rebates.
• At risk: strong history but no purchase for a long time. Highest priority for reactivation.
• Lost: weak in all three attributes. Reactivation only if the effort is covered.

Limits of the method

RFM knows purchase behaviour, not interest. Someone who buys shoes and never looks at clothing is treated as a good customer and bothered with clothing offers. Margin is missing too: a customer with high revenue and a high return rate can be less profitable than a small customer with a strong margin. Anyone using the method in earnest therefore combines it with customer lifetime value and with signals from category interest.

A second reservation concerns the thresholds. Fifths are convenient but arbitrary. In a shop where customers buy twice a year, the resulting groups barely separate anything meaningful. In such cases it is more honest to set boundaries from the observed distribution instead of forcing equal fifths.

How it turns into action

The value appears only when measures are attached, not from the segments themselves. A matrix with segment, trigger, channel, frequency and expected return works well. After that the budget can shift: less waste from newsletters to everyone, more reactivation of the at risk group, and a decision about whether the lost group is addressed at all.

Technically a report over the orders, grouped by customer and period, is enough. A fixed evaluation window, usually twelve months, and a fixed update cadence matter. Otherwise the segments are not comparable from one month to the next and any measurement of success is worthless.