Successful monitoring of machine learning pipelines requires a robust infrastructure for detecting drifting inputs—termed data drift, or drifting target labels—termed concept drift. When performance degradation is observed, the deployed model can be updated.