Model Governance
Processes and controls for the entire lifecycle of ML models: Development, validation, deployment, monitoring, and retirement.
Model Governance controls ML lifecycle: Development → Approval → Deployment → Monitoring → Retirement. Model registry and approval workflows are core.
Explanation
Model governance includes: Model registry (version control), approval workflows, model cards, performance monitoring, drift detection, retraining policies, retirement criteria. Financial sector leading (SR 11-7).
Marketing Relevance
Every production ML needs governance: Which model is running? Who approved it? When will it be retired? Without governance: Chaos and risk.
Example
A model registry stores all versions of the churn prediction model with metadata: Who trained, what data, what performance, who approved for production.
Common Pitfalls
Governance too bureaucratic: Slows innovation. Governance too lax: Uncontrolled risk. Find balance.
Origin & History
Model risk management comes from finance (Fed SR 11-7, 2011). Spread with ML adoption. MLOps platforms (MLflow, Weights & Biases) integrate governance features.
Comparisons & Differences
Model Governance vs. AI Governance
AI Governance is strategic (policies, ethics); Model Governance is operational (registry, workflows, monitoring).
Model Governance vs. MLOps
MLOps is the technical practice; Model Governance is the control framework above it.