RiskOwn ModelsHigh complexity
Proprietary Risk Scoring
Use years of internal loan performance data to price and select risk in ways a bureau score cannot.
How we approach it
Survival and multi-task models (PD, LGD, EAD) trained on internal performance, collections and relationship data. A feature store keeps training and scoring consistent. Training and inference stay on internal infrastructure, and each model sits in the model inventory with validation, challenger models and documented limits.
Business value
Sharper risk selection, on models the bank owns
Technology stack
- PyTorch
- Feast
- MLflow
- internal GPU cluster
- model inventory
Related service
Proprietary AI DevelopmentRelated use cases
- Financial Services · RiskCredit Risk AssessmentScore borrowers on financials, payment history and macroeconomic data, with decisions that a credit committee, a customer and a supervisor can each follow.
- Financial Services · RiskRegulatory Compliance CheckAnswer questions on CRR3, MiFID II, DORA, the AI Act and internal policy, with the article behind every answer.
- Financial Services · RiskFraud Detection & ResponseCatch fraud on instant payments, where money leaves in seconds and authorised push payment scams look like ordinary customer behaviour.