RFM segmentation
RFM stands for Recency, Frequency, Monetary. It is a customer-segmentation method that scores each customer on a 1-to-5 scale across three dimensions: how recently they last bought (R), how often they buy (F), and how much they spend (M). The combined score is a 3-digit code — e.g. 555 (champions) or 111 (lost). RFM was popularised in direct-marketing in the 1990s and remains the cleanest non-ML segmentation method.
How to compute the scores
Use percentile-based bucketing (PERCENTRANK in Excel) so each score 1-5 contains roughly 20% of customers. Equal-width buckets give bad results when data is skewed (which it usually is).
Standard segment archetypes
Champions (555, 554, 545): recent, frequent, big spend — ~5-10% of customers, ~25-40% of revenue. Loyal (453, 543, 444): high F and M, R drifting. At risk (155, 154, 144): high historic value, very low recency — the most important segment to act on. Hibernating (111, 112): low everywhere — suppress from active campaigns.
When RFM works
Transactional businesses with one product line and meaningful repeat purchase — e-commerce, SaaS with usage-based billing, retail. ~6 months of data, ~500+ customers.
When RFM doesn't work
B2B with very few large customers (no statistical mass to bucket). Subscription businesses where everyone has the same Monetary value. Highly seasonal businesses where Recency is dominated by season, not loyalty.
Alternatives
K-means clustering on behavioural features. Predicted CLV. Purpose-built customer-success scoring. RFM is the cheapest and most defensible starting point; the others are upgrades when you have the data and the team.