Model Performance Degradation and Alerting

Why alerting thresholds are set on model metrics.

Why configure alert thresholds on production model metrics?