📊 RFM Segmentation and Churn Prediction

About this simulation

Every point in the cube is a customer placed along three axes: how Recently they bought, how Frequently they buy, and how much Monetary value they generate. Retention teams cluster this cloud into named segments (Champions, Loyal, At Risk, Hibernating, New) and train a gradient-boosted model to output each customer's probability of churning — the same two-stage pipeline described in the article.

How it works

• Segments are assigned from quantile-scored R/F/M values, the classic unsupervised RFM approach marketing teams use before any ML model exists.
• Churn probability is a logistic blend of recency, frequency and monetary signals — a simplified stand-in for an XGBoost classifier trained on real purchase history.
• The translucent risk plane marks the recency point where blended churn score crosses your threshold; customers behind it flip to "predicted churn".
• Firing the retention campaign nudges the highest-value at-risk customers back toward the "active" end of the recency axis, mimicking a win-back offer working.

Controls

• Customers — size of the simulated customer base.
• Recency / Frequency weight — how heavily each factor drives the churn score (monetary takes the remainder).
• Churn threshold — probability cutoff above which a customer is flagged as a predicted churner.
• Color by — switch the point cloud between RFM segment colors and a red-to-green churn-risk gradient.
• 🎯 Launch retention campaign — simulates a win-back push on the riskiest high-value customers. Drag to rotate, scroll to zoom.

Did you know?

Recency is consistently the single strongest predictor of churn across retail and subscription datasets — a customer who hasn't purchased in 90+ days is often a better churn signal than dozens of behavioural features combined.