What's the biggest mistake when rolling out System Monitoring, and how do you avoid an alert flood?

Turning on too many alerts at once → a false-alert flood and the team losing trust in the tool.

Avoid it by starting slow with only the most critical alerts. Also: build a RACI matrix; run one custom baseline template per object type for a few months before subdividing; separate prod/non-prod templates; use counter thresholds for random spikes; and if one metric keeps false-firing while sharing an alert, set that alert's event rule to bestcase.