Customer Churn Prediction

The business question

Keeping a customer almost always costs less than finding a new one. But a team cannot call everyone, so the real question is: which customers are most likely to leave, and who should we contact first?

What I built

• Exploratory analysis of customer behaviour and account data, to find the signals that come before churn
• Feature engineering from that history (tenure, usage and billing patterns)
• Several classification models, trained and compared
• Evaluation with business-relevant metrics: precision and recall on the highest-risk customers, not accuracy alone (if 5% of customers leave, a model that predicts "nobody leaves" is 95% accurate and useless)

How it is used in a business

The model produces a ranked list of at-risk customers for the retention team, with the main reasons for each, so the conversation can address the actual problem.

How success is measured

Of the customers the model flags, how many were actually about to leave; and over time, the churn rate among contacted customers compared with similar customers who were not contacted.