AdInsights operates a real-time personalization endpoint and must detect bias drift weekly using live traffic. The team already enabled data capture to s3://adinsights-capture/. What is the recommended approach to schedule bias checks on the endpoint’s live traffic using SageMaker Model Monitor so that bias metrics run on new data every week?

The correct approach leverages SageMaker Model Monitor, specifically the ModelBiasMonitor type, which is designed for detecting bias drift. By referencing the S3 capture prefix, it directly uses the live traffic data. Setting ScheduleConfig to a weekly cron expression ensures regular, automated checks. The pre-computed bias baseline (constraints/statistics) provides the reference point for comparison, and the IAM role's S3 access is crucial for Model Monitor to retrieve the data. CloudWatch Contributor Insights is for analyzing log patterns, not for comprehensive bias detection against a baseline. SageMaker Clarify explainability focuses on understanding model predictions, not continuous bias monitoring of live traffic in the way Model Monitor does. A Step Functions workflow with Redshift is overly complex and less efficient for this specific task compared to Model Monitor's integrated capabilities.