Let an ad spend three times your target cost per acquisition before you judge it, and up to five times if it has already produced at least one conversion. With zero conversions at three times target, turn it off. With one conversion at three times target, give it room to five times, because a second conversion can pull the average back to target very quickly.
The reason the threshold moves is that one sale changes what you are looking at. An ad with no conversions at three times target is telling you something clear. An ad with one conversion at the same spend is showing you a real result attached to a small sample, and cutting it there means you never find out whether the number was going to settle.
Cheap events need a wider band. For a low-cost event like a lead or a subscribe, the daily swing is much larger, and a cost per lead can move from twenty dollars to five dollars inside a day or two. For those, seven to ten times target is the more honest threshold. Calls and purchases stay at three to five.
There is an earlier signal worth using. If cost per landing page view is already far outside what the account normally produces, you can cut before reaching the spend threshold, because the funnel math will not work regardless of how long you wait. The default when nothing is clear is still to sit it out, because across accounts the far more common mistake is cutting too early rather than waiting too long.