Question

A company runs SageMaker real-time endpoints in multiple AWS accounts (one per business unit). Each account uses SageMaker Model Monitor with data quality and model quality schedules. The central ML platform team needs a single dashboard and unified alerting across all accounts to spot drift and endpoint anomalies, while keeping each account's raw capture data isolated in its own S3 bucket. Which approach best meets these requirements with the least custom code?

Accepted Answer

Configure each account's Model Monitor to publish CloudWatch metrics, then use CloudWatch cross-account observability to share those metrics into a central monitoring account with unified dashboards and alarms.

Answer

Grant the central team cross-account read access to each account's S3 capture bucket and have them re-run Model Monitor baselines centrally against the copied data.

Answer

Replicate all endpoint capture buckets to a central account with S3 Cross-Region Replication and build a QuickSight dashboard on top of the aggregated raw JSON captures.

Answer

Deploy a single shared endpoint in the central account that all business units invoke, so Model Monitor only needs to run once and metrics are naturally centralized.