Datadog APM gains 3 superpowers: App Analytics, Service Map & Watchdog

This article details how Netflix transitioned from relying on unlabeled data and anomaly detection to a supervised learning approach for identifying faulty deployments. Initially, they used metrics to flag unusual behavior, but struggled with false positives. By labeling historical incidents and training a model on that data, they significantly improved the accuracy of detecting real deployment issues and reduced alert fatigue.