Context
In retail lending, declared income is the most critical yet least verifiable input to a credit decision. Requesting documents slows the process and degrades the experience; not requesting them makes the risk invisible. The way out of that dilemma is to estimate income statistically from independent data and compare it against the declaration. Built correctly, the same model is not only a control instrument but a marketing one.
Approach
- The parameter set feeding the model was defined and the reliability of data sources assessed.
- Use cases were separated: marketing target selection, customer segmentation, pre-approved limit setting, accept-reject decisions, declaration validation, and card and overdraft limit management.
- For each use case, how the model output should be interpreted and which thresholds applied were defined.
- The principle was preserved that the model output would not decide alone but be weighed with other risk indicators.
Outcome
- Declared income became comparable against an independent reference without requesting documents.
- Pre-approved limit setting was placed on a data-driven basis.
- The same model became usable for both risk control and targeting.
- The share of intuitive judgement in the credit decision fell.
This case study describes the project through its scope and approach. Client name, commercial figures and performance metrics are withheld under confidentiality obligations.
What This Project Left Behind
A model's value comes less from the estimate it produces than from the clarity of the decision it feeds. The same income estimate needs one threshold when used to set a limit and another when used to decline. It is the first question asked when building a financial model in a company too: which decision will this number change?
- Banking & Credit Process Advisory — Strategic advisory on restructuring distressed loans, financial analysis, negotiation preparation and process management with banks.