RFM scores every customer on three behavioral dimensions: Recency (how recently they last ordered), Frequency (orders in a window), and Monetary (how much they have spent). The combined score places each customer in a lifecycle segment — champions, loyal, at-risk, hibernating, lost — that maps to a specific marketing action. The job is rarely “track a number.” It is “decide what to send this customer right now.”
How the score works
Each dimension is typically scored on a 1–5 quintile, with 5 representing most recent, most frequent, or highest spend in the brand’s own base. Three quintiles produce 125 cells, which operators collapse into 8–12 named segments: R=5 F=5 M=5 is a champion, high-RF with slipping Recency is loyal, R≤2 with F≥3 is at-risk, R=1 F=1 is lost. The naming is convention; the segment-to-action mapping is where the framework earns its keep.
RFM is a methodology, not a metric. “Our RFM went up” is a category error: the output is segment membership, and the trend you watch is segment population over time — champions growing, hibernating shrinking, at-risk recovering after a win-back.
Frequency and Monetary need a defined lookback; 12 months is the common default, longer for considered-purchase categories. Refresh weekly or monthly — slower and segments lag, faster and one order bounces a customer across them. Filter returns and cancellations, or Frequency drifts.
Versus cohort analysis
Cohort analysis groups customers by an acquisition event and tracks that fixed group over time; RFM groups by current behavioral state and re-evaluates on every refresh. A champion in RFM can sit in any acquisition cohort; a March cohort spans every RFM segment by month six. Cohorts answer “how is this vintage performing”; RFM answers “what should we do with this customer this week.” Complementary, not competing.
The use cases follow from the definitions. Win-back targets R≤2 with F≥3. A VIP early-access offer gates on R≥4, F≥4, M≥4. A discount blast suppresses anyone with R=5 who bought in the last 14 days — discounting a customer who just paid full price trains them to wait. Lost customers come off the reactivation budget.
Common pitfalls
The most common pitfall is copying quintile cutoffs from a template. A “frequent” coffee customer orders weekly; a “frequent” mattress customer orders every five years. A 90-day Recency cutoff that flags a coffee buyer as at-risk describes a healthy mattress buyer. Thresholds have to be category-calibrated against the brand’s own purchase-cycle distribution. Computing quintiles on the brand’s own population handles this implicitly — the 80th percentile is the 80th percentile of your customers — the argument for percentile scoring over hardcoded thresholds.
RFM segments correlate strongly with future LTV, which is why most ESPs and CDPs surface them natively. The caveat to keep at the front: Monetary is a revenue signal, not a margin signal. A high-M customer on a thin-margin mix or buying only on discount is not automatically a champion in P&L terms. The segments are starting points for action, not verdicts on customer value.