Intermediate

Model Lifecycle Governance

Implement governance controls across every stage of the model lifecycle, from initial ideation and development through production deployment, ongoing monitoring, and eventual retirement.

Lifecycle Stages

StageGovernance ActivitiesKey Deliverables
IdeationUse case assessment, risk classificationBusiness case, initial risk tier
DevelopmentStandards compliance, documentationModel documentation, test results
ValidationIndependent review, challenger analysisValidation report, findings
DeploymentApproval gates, change managementDeployment approval, rollback plan
ProductionMonitoring, periodic reviewMonitoring reports, review records
RetirementDecommission planning, archivalRetirement approval, archive records
✅
Governance Gate Principle: Each lifecycle transition should have a gate with defined criteria that must be met before proceeding. Gates prevent models from advancing to the next stage without completing required governance activities.

Stage Gate Criteria

  1. Development to Validation Gate

    Complete model documentation, developer testing results, data quality assessment, and initial fairness evaluation before submitting for independent validation.

  2. Validation to Deployment Gate

    Clean validation report with no critical findings, remediation of significant findings, and sign-off from the model risk management function.

  3. Deployment to Production Gate

    Technical deployment readiness, monitoring configuration, rollback procedures, and business owner sign-off on model performance.

  4. Production Review Gate

    Periodic reviews based on risk tier confirming continued model fitness, monitoring health, and alignment with original use case scope.

ML-Specific Lifecycle Considerations

  • Retraining Cycles: Define governance requirements for model retraining - when does a retrained model require re-validation versus automated quality checks
  • Feature Changes: Establish processes for adding, removing, or modifying features that may change model behavior significantly
  • A/B Testing: Governance for running experiments in production, including approval for test design and success criteria
  • Model Versioning: Track lineage across model versions with clear records of what changed and why

Lifecycle Management Tools

Model Registry

Central system tracking model versions, metadata, lifecycle stage, and transition history across the entire portfolio.

Workflow Engine

Automated workflow management for gate reviews, approvals, and notifications with configurable rules by risk tier.

Documentation Platform

Structured documentation templates with auto-population from model artifacts and validation results.

Dashboard

Portfolio-level visibility into model lifecycle stages, upcoming reviews, overdue validations, and governance compliance.

💡
Next Up: In the next lesson, we will design multi-stage approval workflows with role-based gates and automated checks.