How Are You Handling Alert Fatigue at Scale?

I'm reviewing our monitoring strategy for a growing set of services running across multiple environments, and I've noticed that alert fatigue is becoming a bigger issue than actual outages. We have good coverage, but the signal-to-noise ratio isn't where I'd like it to be.

For those managing production workloads, what approaches have worked best for reducing unnecessary alerts without missing critical incidents? I'm particularly interested in practical experiences around threshold tuning, anomaly detection, or alert aggregation.