Facebook ads
Trent Kurth and the host argue that under CBO with cost caps, media buyers should never kill an ad for performance reasons. The old 1x AOV kill rule came from a different account structure, and humans analyzing past ad performance predict future results badly compared with Facebook's algorithm.
The host says that telling the Admission group to 'never ever kill an ad' caused confusion, so this follow-up explains what he meant and why.
The old 1x AOV kill rule came from an earlier era of account structure: all-auto campaigns, many unconsolidated audience-based ad sets, and budgets spent across many placements. Buyers had to police spend manually, which led to killing ads, reviving them after delayed purchases, and 'zombie ads'.
Cost caps and CBO remove the need for manual performance analysis. The buyer hands allocation to Facebook, which is assumed to have better inputs about how to distribute budget across ad sets and ads, and it cuts spend on bad ads itself.
The core claim is that humans are terrible at analyzing data sets and predicting the future, especially by eyeballing columns of data without a calculator. The goal is to remove human decision-making from tactical ad buying as much as possible.
Past performance has no direct causal relationship to future performance. Facebook's serving decisions weigh user behavior and broader data inputs, not just past CTR or cost per result, so a manual kill based on previous impressions gives inaccurate decisions.
'Never kill an ad' applies to performance reasons only. Non-performance reasons remain valid: a sale ended, inventory ran out, or a new angle is being introduced.
Without CBO or cost caps (on auto), the speaker still advises against a quick 1x AOV kill. It is too short a period and too few people (a couple thousand) for the algorithm to learn who to show the ad to, so ads should get through the learning phase, which he puts at roughly 2-3 days.
Ads can be turned off when a new ad gets no spend under a cost cap after about three days, to make room for new ads. The team also recycles old ads, because reintroducing an ad at a new moment is a new experience that can give different outcomes.
Delayed attribution made the old kill-then-revive routine self-defeating. Ads killed at 1x without a purchase would sometimes get a purchase later that day.
Killing at the ad level, sorted by spend, sometimes shut off whole ad sets without the buyer realizing it.
The old 1x AOV rule was based on the team's own research and fit an account structure that no longer exists.
An ad that was outperforming in past impressions may not be the best one to serve to a specific user right now. Facebook may know a user just searched for shoes and serve the shoes ad, or the pants ad, depending on that user's signals.
The ad-set ad count rule loosened from a strict three ads to about six under cost caps, which the speaker says matches Facebook best practice as long as the budget supports it.
An ad getting no spend under a cost cap is treated as the platform's verdict and a signal to swap it out, not as a performance analysis done by the buyer.
The speaker cites Eugene Schwartz to say that each introduction of an ad at a new moment is a new experience, which is why retired ads are worth retesting.
«humans are terrible at making predictions about the future with advanced algorithmic models let alone in their head on the fly»
— 04:06
«never have to make a decision to kill an ad for performance reasons»
— 04:26
«just because an ad performed better in the past does not mean it will perform better in the future»
— 26:27
«manual decision making in the at the ad level based on previous performance is going to lead you to inaccurate decision making»
— 09:09
Reception
Mostly positive and engaged: viewers thank the creator and find the content clarifying, while several ask practical follow-up questions or push back mildly on the 'never kill an ad' advice.
A concise explanation of CTC's shift from manual 1x AOV kill rules to algorithm-led allocation under CBO and cost caps. Its main claim, that past ad performance is a poor guide to future results, is asserted more than demonstrated, and it leaves open how to set the cost cap or handle breakeven targets.

09:29