Scoring and automated decisions
Many lenders decide on applications with rules nobody has tested against data, or by hand. We build a scoring model on your own lending history, document the methodology and embed the model in the process so decisions are made automatically. For online lending we add customer verification through Sumsub.
- A decision on an application takes days, not minutes
- Scoring relies on rules nobody has tested
- Different loan officers decide differently on similar applications
- Arrears are rising and the cause is hard to find in the portfolio
- Standard applications are decided automatically
- Similar applications get the same decision
- The methodology is documented for the auditor and the regulator
- Shifts in portfolio quality become visible sooner
Four steps
Data
We collect the history of applications, loans and repayments, check its quality and agree what counts as a default.
Model
We build the scoring model and test it on a hold-out sample, then set approval cut-offs together with risk and the business.
Integration
We plug the model into the lending flow so application, customer checks, scoring and decision run without manual steps. Borderline cases go to manual review.
Monitoring
We track model performance on new loans and recalibrate it when the portfolio or the market changes.