This project analyzes customer churn for ~10,000 banking customers using SQL for analysis and Power BI for visualization. The objective was to move beyond demographics and identify behavior-driven churn patterns with real business impact.
Churn is not random — it clusters strongly around customer inactivity.
Inactive customers churn at ~36%, compared to ~18% for active customers. This gap holds consistently across age, balance, tenure, and geography, making activity status the strongest churn driver in the dataset.
Age looked like a churn driver — until the data was split by activity. Customers aged 51–65 show high churn overall (~53%), which initially suggests age is the risk factor. But within that same age group, inactive customers churn at ~87% while active customers churn at only ~31% — proving disengagement, not age, is the actual cause.
Geography tells a similar story once volume is accounted for. Germany shows a higher churn rate than France, but France has roughly twice Germany's customer base. Once customer volume is factored in, France's overall churn risk is considerably lower than the raw rate comparison suggests.
Balance and product holding interact with inactivity, not replace it. High-balance customers (₹1L+) churn slightly more (~25%) than lower-balance customers (~21%), but that churn concentrates almost entirely among inactive customers holding only one product — active high-balance customers with 1–2 products stay relatively stable. Customers with a single product churn the most, especially when inactive, while customers with 2 products consistently show the lowest churn, pointing to product diversification as a retention lever.
Some commonly assumed drivers showed weak impact. Gender and estimated salary produced nearly uniform churn (~20%) across segments and were deprioritized in the risk analysis.
Customers don't churn because of who they are. They churn because they stop engaging.
- Prioritize reactivation of inactive customers — activity status is a stronger churn predictor than any demographic variable, so retention spend should target engagement, not age or income segments
- Encourage early multi-product adoption — single-product customers are the highest-risk group; cross-sell timing matters more than the product itself
- Monitor engagement signals over demographics — age and geography only look predictive until inactivity is controlled for; a churn model built on demographics alone would misallocate retention resources
- SQL — data cleaning, segmentation, multi-factor churn analysis
- Power BI — interactive dashboards and churn visualization