Recency-Frequency-Monetary (RFM) Analysis

Measures the time elapsed since a customer's last purchase. The core assumption is that customers who purchased recently are more likely to respond to a new offer than those who purchased long ago.

• Calculation: Days since last transaction date.
• Scoring: Customers are sorted by recency and assigned a quintile score (e.g., 5 = most recent 20%, 1 = oldest 20%).
• Behavioral Insight: A low recency score often signals defection risk, triggering a win-back campaign.

Quantifies the total number of purchases a customer has made within a defined analysis period. It serves as a proxy for engagement and habit formation.

• Calculation: Count of distinct transactions in the observation window.
• Scoring: Customers are ranked by purchase count and assigned a quintile score (e.g., 5 = most frequent 20%).
• Behavioral Insight: High frequency indicates loyalty and is a strong predictor of future repeat purchases.

Represents the total amount of money a customer has spent over the analysis period. It differentiates high-value customers from bargain hunters.

• Calculation: Sum of transaction values or average order value.
• Scoring: Customers are ranked by total spend and assigned a quintile score (e.g., 5 = top spenders).
• Behavioral Insight: High monetary value customers often justify disproportionate retention investment.

Combines individual R, F, and M scores (typically 1-5) to create a 3-digit customer code. This segments the base into 125 distinct behavioral cohorts.

• Example: An RFM 555 customer is a "Champion"—recent, frequent, and high-spending.
• Example: An RFM 511 customer is a "High-Value Lapsed" customer who spent a lot but hasn't returned.
• Actionability: Each segment receives a tailored marketing strategy, from loyalty rewards to reactivation discounts.

RFM analysis relies exclusively on structured transactional logs. It does not require demographic, psychographic, or clickstream data, making it universally applicable where purchase records exist.

• Required Fields: Customer ID, transaction date, and transaction amount.
• Data Quality: Accuracy depends on deduplicated, clean transaction records.
• Limitation: It ignores browsing behavior and product affinity, making it a lagging indicator of intent.

Traditional RFM is a batch-processed, historical snapshot. Modern implementations evolve this into a dynamic model updated in near real-time.

• Static RFM: Scores calculated weekly or monthly; suitable for email campaigns.
• Dynamic RFM: Scores recalculated on every transaction event using Event Stream Processing (ESP).
• Advantage: Dynamic RFM enables immediate triggers, such as a "save the sale" offer when a high-value customer's recency score drops.

A statistical technique that calculates the probability of a specific future action, such as a purchase or churn. It directly operationalizes the 'Recency' dimension of RFM by quantifying the likelihood of conversion based on a user's current digital body language.

• Converts raw behavioral signals into a single actionable probability.
• Used to trigger real-time interventions, like a discount offer for a user with a high churn propensity.
• Models are typically trained on binary classification algorithms like logistic regression or gradient-boosted trees.

The practice of dividing a customer base into extremely granular, highly specific groups. It moves beyond the static 3x3x3 RFM grid to create dynamic, intent-based clusters.

• RFM creates 125 segments (5x5x5).
• Micro-segmentation creates thousands of segments by adding real-time behavioral, demographic, and contextual data.
• Enables 1:1 personalization by treating each segment as a distinct persona with a unique next-best-action.

A stream processing operation that continuously computes summary statistics over a finite, time-bounded subset of an event stream. This is the technical mechanism that transforms static RFM into real-time RFM.

• A sliding window computes a user's purchase frequency over the last 30 days, updating with every new transaction.
• A session window groups all events within a single user visit to calculate real-time monetary value.
• Frameworks like Apache Flink and Kafka Streams provide native support for these operations.

A metric that quantifies the strength of a user's preference for a specific product, brand, or category. It adds a critical qualitative dimension to the quantitative RFM model.

• RFM tells you how a customer buys.
• Affinity Scoring tells you what they love, based on engagement history and behavioral signals.
• A high-monetary-value customer with a strong affinity for a niche category should receive different recommendations than one with broad, generic interests.

The process of monitoring and identifying when the statistical properties of a target variable change over time. In an RFM context, this detects when the underlying purchasing behavior of a segment has fundamentally shifted.

• A customer's 'Frequency' may drop not due to disengagement, but because of a seasonal pattern or a life-stage change.
• Drift detection algorithms trigger alerts to retrain models or re-evaluate static segmentation thresholds.
• Prevents marketing automation from acting on stale, inaccurate segment assignments.