By Jonas Osman AbdelfourPublished February 18, 2026
Summary Independent model validation is a control over model risk. Its purpose is not to reproduce model development, but to test whether the model is fit for the decisions it supports.
Scope of validation Validation covers conceptual soundness, data quality and lineage, methodology, implementation, performance and the surrounding governance and controls.
Testing techniques Benchmarking, sensitivity and stability analysis, out-of-time and out-of-sample testing, and — where feasible — replication of the model.
Governance of findings Findings are rated, prioritised and mapped to remediation. Compensating controls are identified where remediation is not immediate.
AI and machine-learning models The same standard applies, with additional emphasis on data drift, feature stability, explainability where it affects decisions, and monitoring of performance in production.
Related expertise See [Model Risk and Validation](/expertise/model-risk).
Frequently asked questions
What should risk leaders know about scope of validation?
Validation covers conceptual soundness, data quality and lineage, methodology, implementation, performance and the surrounding governance and controls.
What should risk leaders know about testing techniques?
Benchmarking, sensitivity and stability analysis, out-of-time and out-of-sample testing, and — where feasible — replication of the model.
What should risk leaders know about governance of findings?
Findings are rated, prioritised and mapped to remediation. Compensating controls are identified where remediation is not immediate.
What should risk leaders know about aI and machine-learning models?
The same standard applies, with additional emphasis on data drift, feature stability, explainability where it affects decisions, and monitoring of performance in production.