Twenty-four of last month's sixty-three incidents were suppressed — reviewed, matched against known activity in that environment, closed, and tuned so the same benign pattern wouldn't raise a second one. Thirty-eight percent of the queue.
Here is the clearest example. One alert type — a new local user account created on a Windows machine — fired eleven times last month, on eleven different computers. Ten were technicians imaging hardware, setting up a kiosk, or standing up a service account, and were closed as expected. One wasn't, and was worked as a real investigation.
If you had bought that detection and pointed it at your own inbox, you would have received eleven identical alerts, ignored the first four, and been trained by the fifth to ignore the eleventh. That is precisely how the alert that mattered gets missed — not through absence, but through volume.
Deciding which of the eleven was different is the entire product. It requires knowing what that environment's technicians were doing that week, and it cannot be done by a rule.