Customer Churn
Prediction

Customers Were Leaving — But Why?

The company was experiencing a steady increase in customer churn, but the team was only discovering lost customers after they'd already left. There was no system in place to flag at-risk accounts early enough to intervene.

Retention efforts were reactive and unfocused — reaching out to everyone equally instead of prioritizing the customers most likely to leave. This wasted resources and missed the window for effective intervention.

Data-Driven Early Warning System

I designed a predictive framework that combined historical behavior with leading indicators:

• Data Extraction: Pulled 24 months of customer interaction data from the SQL database — purchase frequency, support tickets, engagement scores, and contract details.
• Feature Engineering: Using Python, created behavioral features such as declining purchase trends, increasing support contact frequency, and days since last engagement.
• Risk Scoring: Built a scoring model that assigned churn probability to each active customer based on their behavioral patterns compared to historical churners.
• Actionable Output: Delivered a prioritized Excel report to the retention team, highlighting the top at-risk accounts with specific reasons for their elevated risk.

Engagement Drop Preceded Churn by 6 Weeks

The most powerful predictor wasn't complaints or pricing — it was a quiet decline in engagement. Customers who reduced their interaction frequency by 40% or more had a 78% likelihood of churning within 6 weeks.

This insight gave the retention team a critical window: instead of waiting for cancellation requests, they could proactively reach out with personalized offers during the "silent decline" period — when intervention was most effective.