Customer churn prediction

Problem

Annuity retention needed a ranked list of policyholders likely to leave, not another dashboard of averages.

Constraints

• Sample of roughly 100,000 customers
• Operations needed a list they could act on, not a research paper
• Compare classical ensembles against a small neural net honestly

What shipped

Random Forest, Gradient Boosting, and a small neural net trained on the same cohort. Random Forest won on 92% accuracy and 0.85 F1 and became the retention team’s targeting list. Age, income, and occupation dominated feature importance.

Outcome

A production targeting input for retention outreach — one number the team could defend in review.