Deploying an AI model is the beginning of the lifecycle, not the end.
Why ongoing monitoring matters
AI models:
- learn from changing data
- operate in evolving environments
- degrade over time
Without monitoring, performance and fairness can erode silently.
Detecting drift responsibly
Operating models should define:
- performance thresholds
- drift indicators
- review cadence
Drift detection is a governance function, not just a technical one.
Knowing when to retrain or retire models
Not every model should be retrained indefinitely.
Institutions must decide:
- when retraining is appropriate
- when replacement is safer
- when retirement is required
Undocumented models lingering in production create risk.
Retirement is part of governance
Model retirement should include:
- decommissioning controls
- evidence retention
- replacement planning
A model that cannot be retired cleanly was never governed properly.
Read next: → Operating Models for Regulated AI