The best practice is to test and modify as needed. It's very likely that the first set-up for your alerts will need to be modified at least once, especially if you don't have a lot of historical data captured yet. Also, getting the sensitivity right is important. Something might be an anomaly - but is it worth looking into? It depends on your individual business, so it'll take some time to find that right level. If you get alerts too frequently, people will stop paying attention to them. If you don't get them often enough, you might miss some important events.
Other than that, it's really going to depend heavily on your website traffic. Do you see big fluctuations based on seasonality? Then it might be better to use the 'anomaly exists' with 90/95/99% thresholds instead of static numbers that you would have to update with each new season. But if your traffic is stable year-round, then static numbers could be the way to go.
You can also use relative metrics instead of absolute. For example, instead of saying orders is below X amount, measure conversion rate. Or measure ATC rate instead of actual cart adds. Relative metrics can tell you a very different story than absolute numbers. You can also set up alerts for both types of metrics.
The other suggestion I have is to set up segments to exclude time periods you don't want to get alerts for. If you set up an hourly alert, it's going to check every hour for an anomaly. But if your traffic drops overnight (when everyone is asleep), it's more likely you'll get false positives. What I've done is make segments that exclude the hours between 11pm and 5am, that way we don't get false positives.
