1. Assessment

  • Triage SageMaker spend and define scan scope
  • Inventory persistent SageMaker resources by region
  • Shortlist suspect persistent resources
  • Check targeted CloudWatch metrics for shortlisted resources
  • Deep inspect remaining suspect resources

2. Summary

  • Produce optimized unused SageMaker resources report
1 Credits

Identify SageMaker Unused Resources and Savings Opportunities

Overview

Review Amazon SageMaker spend and identify potential savings opportunities from unused or forgotten resources. This plan uses recent Cost Explorer spend, limits detailed inspection to spend-bearing Regions, inventories persistent SageMaker resources, checks targeted CloudWatch metrics and logs where useful, and produces an evidence-based summary of resources that may be safe candidates for cleanup or deeper owner review.

Execution Details

1. Assessment

  • Triage SageMaker spend and define scan scope
    Confirms recent SageMaker spend, identifies spend-bearing Regions, and determines whether resource-level attribution is available before scanning resources.

  • Inventory persistent SageMaker resources by Region
    Builds a bounded inventory of persistent SageMaker resources such as endpoints, notebook instances, Studio domains, apps, users, and spaces in Regions with spend.

  • Shortlist suspect persistent resources
    Uses low-cost inventory fields such as status, age, last modified time, and Region spend context to identify resources that may be forgotten or underused.

  • Check targeted CloudWatch metrics for shortlisted resources
    Uses SageMaker endpoint metrics and validated resource-specific metrics where available to distinguish active resources from potentially inactive ones.

  • Deep inspect remaining suspect resources
    Performs describe calls and targeted CloudWatch Logs checks only for the reduced suspect set to gather stronger evidence of activity or inactivity.

2. Summary

  • Produce optimized unused SageMaker resources report
    Summarizes spend, scanned resources, potentially unused resources, confidence, evidence, likely cost impact, and uncertainty from missing attribution or unavailable metrics/logs.