Ideas

Next to creating alerts based on fluctuations and thresholds, this third option in Performance Alerts would simply ask for a sensitivity threshold and auto-alert you whenever a metric goes outside its predicted range (boundary), given the historical performance. You wouldn't need to enter deviation percentages or time frames to compare. Machine learning will use models to best predict what should have been the value for a certain metric vs. the actual value.

See the graph below for an example. The outliers (red dots) would trigger an alert.