ML Customer Churn Prediction for 100,000+ B2B Accounts
Challenge
A U.S. storage and records company with 100,000+ corporate customers kept learning about lost accounts after the fact. Survey scores, support tickets, invoice credits, recurring revenue, work order delivery and price revisions all sat in separate systems and had never been analyzed together. Slow payments, falling order volumes and rising complaints only became visible once an account was already gone.
A U.S. storage and records company with 100,000+ corporate customers kept learning about lost accounts after the fact. Survey scores, support tickets, invoice credits, recurring revenue, work order delivery and price revisions all sat in separate systems and had never been analyzed together. Slow payments, falling order volumes and rising complaints only became visible once an account was already gone.
Solution
We combined six data sources on SQL Server 2019 and Microsoft Fabric and trained machine learning models that score every account as Low, Medium or High churn risk, weighted by revenue. The output is a live intelligence layer that sends alerts, with context and a suggested fix, straight to the responsible VP. Operations, Sales and Support now act weeks before a cancellation notice instead of after it.
We combined six data sources on SQL Server 2019 and Microsoft Fabric and trained machine learning models that score every account as Low, Medium or High churn risk, weighted by revenue. The output is a live intelligence layer that sends alerts, with context and a suggested fix, straight to the responsible VP. Operations, Sales and Support now act weeks before a cancellation notice instead of after it.
Results
Revenue retention across large and medium accounts reached 98.5%. Six data sources now feed one view of customer health, and every account sits in one of three risk tiers. The biggest finding surprised leadership: most attrition was driven by operations, not pricing. That insight led to five targeted programs, including annual price revision policies, proactive credits, bundled pricing and seasonal re-engagement campaigns.
Revenue retention across large and medium accounts reached 98.5%. Six data sources now feed one view of customer health, and every account sits in one of three risk tiers. The biggest finding surprised leadership: most attrition was driven by operations, not pricing. That insight led to five targeted programs, including annual price revision policies, proactive credits, bundled pricing and seasonal re-engagement campaigns.