Alert Actionability
The proportion of alerts that result in a human taking action, used as the primary quality measure of an alerting system.
The single most useful number about an alerting system, and one almost nobody tracks.
If most alerts require no action, responders learn — rationally — to assume the next one does not either. That adaptation is the mechanism behind delayed responses to genuine incidents, and it cannot be fixed by telling people to try harder.
Interpretation: above roughly 80% is healthy. Below 50% means the system is training people to ignore it, and every additional rule makes it worse.
Measuring it requires only that responders record an outcome when resolving a page — action taken, no action needed, duplicate, false positive. That small amount of discipline turns alert quality from an opinion into data, and it makes deletion arguments easy to win.
What to do with the result: delete non-actionable rules rather than tuning them; convert conditions that matter but are not urgent into tickets; group related alerts so one cause produces one page; move from cause-based rules to symptom-based ones; and use burn-rate windows so severity tracks impact.
The cultural rule that sustains it: a page requiring no action is a defect, filed and fixed like any other bug — not tolerated as the cost of monitoring.