Stockertim Dictionary · Artificial intelligence · Finance
Model validation
Also called validation of models, independent model validation
In short
Model validation is independent testing that a model works as intended and is fit for its use, carried out before the model is used and repeated during use.
- Intermediate
- Process
- Checked
In detail
Model validation is the set of checks that gives a firm confidence a model does what it is meant to do. It covers whether the design and data suit the purpose, whether results are accurate on data the model has not seen, how the model behaves at the edges, and whether its limits are documented.
Validation is done by people independent of the builders, so that the same assumptions are not marked by those who made them. For a predictive model this means back-testing estimates against outcomes, testing stability and checking for unfair results across groups. For a generative tool it means test sets of realistic tasks, scored against a rubric for accuracy, completeness and faithfulness to sources.
It matters because models fail quietly. A model can perform well in development and poorly in production, or be used for a purpose it was never built for. Supervisors in many countries expect firms to validate models in proportion to their risk, and to revalidate when the model, its data or its use changes, including when a supplier updates an underlying AI model.
Validation findings are recorded, and any limits of use are written into the approval.
Worked example
Kite & Co Accountants validates an AI tool that summarises client contracts before allowing its wider use.
- Reviewers prepare 500 test contracts with known answers.
- The tool gets 470 summaries right on the key terms.
- Accuracy on the test set: 470 / 500 = 94%
- The 30 failures cluster in contracts with handwritten amendments, so the approval excludes those.
The tool is approved for typed contracts only, with a revalidation date and a sample check each month.
In practice
Banks, insurers and asset managers run validation teams within model risk management; smaller firms often use an independent internal reviewer or an outside specialist.
Watch out
- Validation is not a one-off sign-off; it is repeated when the model, data or use changes.
- Testing by the builders alone is development testing, not independent validation.
Learn it properly
More lessons about model validation
Official sources
- AI Risk Management Framework · National Institute of Standards and Technology
- Artificial intelligence in UK financial services 2024 · Bank of England
- NIST AI RMF Playbook · National Institute of Standards and Technology
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.