Answer:
- Start with enabling analysis only for key components (e.g. Lambda + API Gateway) to limit cost and noise
- Ensure your application generates sufficient metrics, logs, and trace data for ML models to establish a baseline
- Keep function versions and performance stable initially, then introduce changes to test detection
- Use DevOps Guru recommendations as guidance, not absolute commands
- Keep traditional monitoring/alerting (CloudWatch alarms) in parallel, especially for critical thresholds
- Review and prune alerts/insights over time to reduce noise
