How an optimisation event works
Once an ad set knows where the action happens, it needs to know which action. The optimisation event is chosen from the events Meta can actually see: standard pixel events such as content viewed, add to cart, checkout started, lead and purchase; custom conversions you have defined from a URL or a rule; app events sent by an SDK; or offline events uploaded from a CRM.
Delivery then builds its picture of a buyer from the people who completed that event, and goes looking for more of them. Everything downstream is expressed in the same terms — the cost per result in your report is the cost of that event, not of a sale, unless the two happen to be the same thing. On the web there is a further constraint: only a capped list of events per verified domain is usable for optimisation and reporting, which is what event prioritisation exists to manage.
Why the optimisation event matters
It is the definition of success that the machine actually uses, as opposed to the one in the proposal. If the event you chose fires on a page anybody can reach, delivery will efficiently find people who reach that page and nothing more. The account is not misbehaving; it is doing precisely what it was asked.
Depth is the trade-off. A deep event such as a purchase describes a real customer but happens rarely, so delivery has little to learn from. A shallow event such as a page view happens constantly but describes almost nobody worth having. The skill is picking the deepest event that still occurs steadily enough for the system to find a pattern in it.
Common mistakes with optimisation events
The most damaging is an event that fires when it should not. A thank-you page event that re-fires on refresh, a lead event dropped on every page of the site, or a custom conversion built from a URL rule that matches more pages than intended will all report success that never happened, and delivery will chase the phantom.
The second is choosing a rare event on a small budget. If the chosen action happens only occasionally, the ad set stays in the learning phase and results swing wildly. The third is renaming or restructuring events without checking what depends on them — audiences, custom conversions and the prioritised list all break quietly, and the campaign carries on spending while reporting nothing.
How to act on it
Verify before you optimise. Open the event in Events Manager, complete the action yourself, and confirm it registers once, with the right parameters, and only on the page or screen it belongs to. An event nobody has tested is not a measurement, it is an assumption.
Then choose the deepest event with steady volume, and treat a shallower event as a temporary stepping stone rather than a permanent redefinition of success — keep reporting the outcome the business cares about even while delivery optimises on something earlier. When you change what you optimise for, revisit the prioritised web event list so the new choice is actually usable, and keep browser and server events matched so the pixel and Conversions API setup counts each action once.
