RFM Analysis

Scoring customers on how recently, how often and how much they bought.

Also known as: RFM, RFM segmentation, recency frequency monetary

RFM analysis scores each customer on three behaviours: Recency (how recently they bought), Frequency (how often), and Monetary (how much they spend). Rank customers into bands on each — typically quintiles — and the combined score segments the base into champions, loyalists, at-risk, hibernating and lost, each with an obvious next action.

customer:  R=5 (bought yesterday)  F=2 (rarely)  M=4 (big tickets)
segment:   "new big spender" → onboard well, second-purchase nudge

Its power is actionability: unlike abstract clusters, RFM segments read as instructions. High-recency low-frequency gets a second-purchase campaign; low-recency high-frequency gets a win-back offer; low everything gets suppressed from expensive channels.

The classic mistakes:

Arbitrary quintiles treated as truth. Band edges are choices, not discoveries. Two customers either side of a cutoff are nearly identical — do not brief them as different species.
Stale scores. RFM decays daily; a monthly refresh lets “at-risk” customers churn before anyone acts. Refresh on the cadence of the action it drives.
Monetary without margin. Ranking by revenue promotes discount-chasers and unprofitable accounts. Score on margin or contribution where it matters.
One RFM for unlike businesses. Subscriptions (tenure matters), marketplaces (both sides matter) and B2B (accounts, not users) all bend the frame. Adapt the letters to the business before segmenting.
Segments nobody uses. Five beautiful segments with no channel, offer or owner attached are wall art. Each segment ships with its action.

When to go further: when behaviour needs more dimensions than three letters — then segmentation with clustering, or cohort analysis for time-based behaviour. RFM is the fast, explainable first cut.