Challenge
A Bangladesh financial institution needed to predict loan defaults before they happened and flag high-risk accounts across a portfolio of 22,000+ loans spanning 14 branches.
Approach
We built a machine learning pipeline using CatBoost for classification, engineered 40+ features from historical loan data, and deployed an Early Warning System that scores every active loan daily.
The system integrates with the institution's existing core banking via API and surfaces results through executive dashboards built on Power BI.
Outcome
Achieved 84.
12% AUC on the prediction model. The system monitors 17,652 active loans in real-time, flagged 1,882 critical-risk accounts in its first quarter, and enabled proactive intervention that measurably reduced default rates.