Same goes for log output - if you get thousands of lines every day, no-one ever looks at the logs and things get missed. If you only get log messages when things are really broken, people actually notice.

I mean, this is what log levels are for. My personal standard is log level error is for things raising an alert over that you need a human to look at, warning is for stuff that can raise an alert if it goes over a threshold, info is only for looking at when a problem has already been identified, so that you can try understand the lead up to the problem, and debug is incredibly spammy stuff only turned on when the above gives insufficient details to recreate the problem locally.

Stats are good for things that may be a problem or may not. It’s hard to get a sense of rate with log analysis. You can, but comparing several stats to each other at once is painful.