In 2013, Jordan Hochenbaum and I developed an automated anomaly detection algorithm for Twitter. Specifically, we were attempting to automate the detection of anomalies within user metrics that exhibit a heavy seasonal component. We worked with Arun Kejariwal, and developed a method called S-H-ESD that removes the seasonal components and then robustly detects outliers in the residual. We presented the work at USINEX HotCloud 2014, and published a paper on the detection of anomalies in long term time series. The project has been open-sourced as an R package, and is available on Twitter's github. The following video shows Arun presenting the work and it's integration into Twitter's anomaly detection system.