Predictive churn modelling for a leading US retail group

The Challenge

Churn was costing the business millions a year, and the retention team had no way to see it coming. By the time a customer was flagged as lost, they were already gone. Re-engagement campaigns existed, but they were expensive and landed too late to change much. The harder problem was that there was no way to tell which customers were genuinely at risk from those who were just in a longer buying cycle.

Our Approach

A churn model was built at the individual customer level, drawing on purchase history, engagement signals, and recency, frequency, and monetary value features. Every active customer gets a churn probability score, updated weekly.

Customers were grouped into risk tiers — high, medium, low — so the CRM team could run different interventions for different segments rather than a single blanket campaign. A control group was kept to measure whether the interventions were actually moving the needle.

The model was connected directly into the CRM so retention workflows could trigger automatically when a customer's score crossed a threshold — no manual reporting step required.

Key Deliverables

• Customer-level churn prediction model with weekly scoring
• RFM segmentation and risk-tier framework (high / medium / low)
• CRM integration for automated trigger-based retention campaigns
• Control group methodology for measuring incremental retention lift
• Ongoing model performance monitoring and accuracy reporting

Results

Engagement

Facing a similar challenge?

Tell us about your data problem. We'll help you find the right path forward.