Direct answer: AI improves RFM mainly by replacing the quintile grid with a probability model that scores each customer individually.
Classic RFM: days since last purchase, total purchases, and total spend, each cut into tie-aware quintiles. It predicts expected holdout purchases for each customer. Target: actual purchases in the holdout.
Key takeaways
- Where the model earns its place Same cell, different customers The chart shows the model’s predicted holdout purchases for customers in the two top…
- Dormancy as a probability: BG/NBD gives each customer a probability of still being active.
- Expected purchases in real units The model predicted 1,770 holdout purchases against 1,882 actual, about 6% low.
Check the spend model’s assumption.
Read the full article: https://www.aipersonalization.cloud/rfm-2-0/
Top comments (0)