BSFI
Predicting Delinquencies of a Personal Loan Portfolio Ahead of Time
75%+ accuracy in predicting loan delinquencies
75% to 82%
overall model accuracy
False Positive
rate of 2% to 6%.
False Negative
rate of 12% to 15%.
OOS Score
of 84%, indicating strong model performance.
Why this Global Bank Loves nanoBI
- Automated data acquisition and reconciliation.
- Machine learning-driven matching and validation.
- Self-serve analytics for better insights.
- Reduction in manual processing time and costs.
75% to 82%
overall model accuracy.
False Positive
rate of 2% to 6%.
False Negative
rate of 12% to 15%.
OOS Score
of 84%, indicating strong model performance.
Loved it? Ready to try nanoBI?
Why this Fintech Loves nanoBI
- An early warning system for delinquency prediction helps this fintech take timely mitigation strategies.
- Uses macro and external variables for better accuracy.
- Predictive insights for proactive risk mitigation.
- Extensive variable library with 90+ factors for accurate scoring.
About the client
A new-age fintech company operating in the BFSI sector in India, managing a diverse personal loan portfolio. The client focuses on data-driven risk assessment and aims to enhance loan performance by predicting delinquencies well in advance.
Challenges Before nanoBI
- No clear visibility into which loans might show stress 6 to 9 months ahead.
- Existing models only used account-related variables, limiting predictive power.
- Needed a better tagging system to classify delinquency risk and take timely action.
nanoBI Solution
- nanoBI’s Early Warning Solution was implemented and tuned to meet the specific segment portfolio.
- The solution scores all live accounts in the specified portfolio every month of their propensity to default or be delinquent over the coming six to nine months.
- The Early Warning Solution uses macro or external variables other than account or internal variables.
- 32+ variables (internal and external) were used to predict early warning delinquencies (the solution has a library of 90+ variables).
How Things Have Changed
- nanoBI’s Early Warning Solution has helped the client achieve an overall model accuracy of 75% to 82%, significantly improving delinquency predictions.
- The false positive rate was kept as low as 2% to 6%, ensuring minimal incorrect flags, while the false negative rate remained within 12% to 15%, balancing risk prediction.
- The model’s OOS (Out-of-Sample) score of 84% further demonstrated its robustness in handling real-world scenarios.
- With early tagging of delinquent accounts, the client can now proactively apply mitigation strategies, reducing risk exposure and improving loan performance.
