Almost every team we work with already has monitoring. Almost none of them trust it. That gap is usually the real problem, not the tooling.
It happens gradually. An alert fires for something that turns out not to matter, so it gets acknowledged and forgotten. It happens again. Eventually the on-call rotation learns, correctly, that most pages aren't worth waking up for — and then the one page that is worth waking up for gets the same treatment.
Fixing this isn't about adding more alerts. Usually it's the opposite: removing most of what currently pages a human, keeping only the handful of signals that reliably correlate with something a person actually needs to act on, and routing everything else to a dashboard instead of a phone.
A rough rule we use: if an alert has fired more than a couple of times without leading to a real action, it shouldn't be an alert. It should be a line on a dashboard, or it should be deleted.
The result isn't fewer things being watched — it's fewer things interrupting a human, which is what actually determines whether an on-call rotation holds up over time.