Model Risk Management
The discipline of identifying, measuring, monitoring, and mitigating risks arising from AI and statistical models used in business decision-making.
Model Risk Management (MRM) is the risk management discipline specific to models — the quantitative and AI tools that organisations use to make business decisions. It originated in financial services following regulatory guidance (US Federal Reserve SR 11-7, 2011) and has since expanded to cover AI models across industries.
Model risk sources:
Incorrect model outputs — Model makes errors due to poor specification, data quality issues, or overfitting. Risk increases as business reliance on the model grows.
Misuse — Model is used for purposes it was not designed for, or in market conditions it was not validated against.
Model drift — Model performance degrades over time as the world changes. This is the most common failure mode for deployed models.
Explanation failures — Model decisions are not accurately explainable, creating legal and regulatory risk.
The MRM framework (SR 11-7): Development and implementation → model validation → ongoing monitoring → model risk reporting → model inventory management. This cycle applies to all material models and is reviewed annually by internal audit.
AI-specific MRM challenges: Traditional MRM was developed for statistical models (logistic regression, linear models) with interpretable parameters. Deep learning and ensemble models require additional tooling — explainability methods (SHAP/LIME) and distribution monitoring (PSI) — to satisfy the same governance requirements.