01 / PUT IT INTO PRACTICE
Connect the result to its evidence
Retain the source version, transformations, feature definitions, model version and policy version. A plain-language explanation helps the reviewer understand the decision; the underlying lineage helps a specialist reproduce it.
02 / PUT IT INTO PRACTICE
Review a model separately from its policy
The model estimates an outcome. Policy determines the permitted action, exceptions and review requirements. A change to an approval threshold is a policy change even if the fitted model stays the same. Both need their own recorded rationale.
03 / PUT IT INTO PRACTICE
Check the fit to your borrowers
Before a model becomes part of your lending process, examine how its estimates compare with outcomes in your own relevant loan history. Check the products and borrower groups it covers, the information it relies on and the situations where a person should review the answer.
- Agree which lending decisions, products and borrowers the analysis will support.
- Keep the performance review, known limitations and missing information together.
- Record who approved the approach and why it is suitable for the decision.
- Set review dates, warning signs and a way to return to the previous approach if needed.
04 / PUT IT INTO PRACTICE
Keep checking as your portfolio changes
Track whether inputs, populations and observed outcomes change. Mature outcomes are needed to evaluate predictive performance; early data checks are not a substitute. Investigate changes before automatically refitting or replacing an approved version.
EXAMPLE / PUTTING IT INTO PRACTICE
A renewal decision is challenged months later. A useful review pack identifies the data available on that date, the behavioural model and policy used, the assumptions and any recorded override. Recomputing with today’s model would answer a different question.
WHAT YOU TAKE FORWARD