miatzSign in

Scoring a loan application against past repayment outcomes

Fits a model to the outcomes of past advances — repaid, rolled, written off — and returns for a new applicant a probability of default together with the features that moved it, which a credit officer applies inside a written policy that still owns the cut-offs, the overrides and the list of who may exercise them.

Effort
More than a few weeks
Skill level
Engineering skill required
Organisation size
Mid-market
Value
Time saved, Risk reduced

Tools named for this

  • A logistic or gradient-boosted model over the lender's own repayment history
  • A reason-code layer stating which features moved a particular score
  • A champion-challenger harness keeping the incumbent policy scoring alongside the new one

What to check before you ship it in India

  • The score drives a decision about an identified person, which engages section 8(3): the Data Fiduciary must ensure the personal data is complete, accurate and consistent where it is likely to be used to make a decision affecting the Data Principal. A stale bureau record or a mis-keyed income is therefore the lender's problem, not the model's.
  • Where the lending is digital, the Reserve Bank of India (Digital Lending) Directions, 2025 require the Regulated Entity to obtain the borrower's economic profile — age, occupation, income — before extending the loan, so that creditworthiness is assessed in an auditable way. A score whose inputs cannot be reconstructed as they stood on the decision date is not auditable however well it discriminates. A lender outside the Regulated Entity perimeter is not reached by this.

Sources

Every claim on this page traces to one of these, on the date it was read.