When data goes wrong, it’s critical to get the context fast. In case of data anomalies, we’ve introduced a preview that we send directly to Slack or email so you get a sense of the magnitude and direction of the shift.

Feedback for models #

When data anomaly happens, besides alerting you, we immediately isolate the anomalous data from metrics we use to train the model to ensure that detected anomalies will not impact the projected data in the future.

Correcting detected anomaly as expected.

You can correct our prediction for cases when you don’t want such behavior—most likely when a detected anomaly is expected. It will resolve the monitor’s status back to success and adjust the model to account for the data shift in the future, not repeating the similar pattern in your metric as an anomaly.