Many problems announce themselves as something out of the ordinary: a sudden spike, an unusual transaction, a reading that breaks a trend. Anomaly detection learns what normal looks like and flags what deviates, often without being told in advance what a problem looks like.
The difficulty is the line between rare-but-fine and genuinely wrong, since too sensitive a system cries wolf and too dull a one misses real trouble. It is most valuable where problems are rare and hard to specify but costly to overlook.