Segment customers with RFM analysis

RFM analysis is the most widely used customer segmentation method in direct marketing and e-commerce, because it is simple, needs only order data and predicts future behaviour remarkably well. Customers who bought recently, buy often and spend a lot are the most likely to buy again.

The three dimensions

• Recency (R): days since the last purchase, measured at an analysis date. Lower is better.
• Frequency (F): number of orders in the period.
• Monetary value (M): total (or average) spend.

Each customer receives a score from 1 to 5 on each dimension, usually by quintiles: the 20% most recent buyers get R = 5, the 20% least recent R = 1. A customer scored 555 bought recently, often and for a lot; 111 is the opposite. Recency is usually the strongest predictor of the next purchase.

From scores to segments

With 125 possible score combinations, the tool groups customers into named segments based on R and F (with M for value):

Reading the result

Champions are 27.9% of customers but generate 55.8% of revenue, a typical Pareto pattern. The most urgent group is not the lost customers but the At risk and Can’t lose them segments: they used to buy often or spend a lot (5.5 and 8.5 orders on average) but have not ordered for over a year. Together they account for 18.8% of past revenue. Win them back before they become lost.

An action for every segment

• Champions: reward them, ask for reviews and referrals, give early access. Avoid unnecessary discounts; they buy anyway.
• Loyal customers: upsell, loyalty programmes, personalised recommendations.
• Potential loyalists and new customers: onboarding, a reason for the second and third purchase, bundles.
• At risk and Can’t lose them: personal win-back outreach, strong offers, and find out why they stopped.
• Hibernating and Lost: low-cost reminders or a test discount; do not spend much.

Choosing the settings

The analysis date should be the day after the last order in the data (or today). The period should cover several purchase cycles: for groceries a few months, for furniture several years. Quintiles adapt to your data but shift over time; fixed business thresholds (for example, “active” means a purchase in the last 90 days) are easier to track month to month. If a few customers have huge frequencies, rank-based scores handle them well.

Measure the effect

RFM tells you whom to target, not whether the campaign works. Keep a random holdout group in each segment that does not receive the campaign, and compare their purchases with the targeted group’s. Without a holdout, a win-back campaign aimed at customers who would have come back anyway looks more successful than it is.

Reporting

RFM analysis of 3,775 orders from 520 customers identified eight segments. Champions (27.9% of customers) generated 55.8% of revenue. At-risk and can’t-lose customers, who had purchased frequently but not in the last year, accounted for 18.8% of past revenue and were prioritised for a win-back campaign with a 10% random holdout.