Creative is the targeting. Broad audience, sharp ad, a new pack every week.
The idea
We don’t pick interests. We give Meta a broad audience and a purchase objective, and the ad itself tells the algorithm who to find. Name the person in the first line or the first frame, and Meta goes looking for more people like them.
So the work goes into the ads. Before we make one, four questions decide what kind it should be: how aware the customer is, what kind of product it is, what stops the sale, and how much budget there is to test with.
Every ad stands on at least one principle that makes people stop and believe: a pattern interrupt, a curiosity gap, loss aversion, authority, specificity or identity.
The account
A starting split. It moves with the numbers.
Broad audience, purchase objective, Audience Network off. Each pack is one avatar and one concept.
By depth of interest, with existing customers kept separate.
Creative
A variation is a new hook or a new shot, never one word changed in the headline.
Account data says what to test. Concepts, brief and copy get written.
AI production, filming and creator videos.
Review, then live as a new pack.
Variations of the winners. The rest fade out.
The rules
The algorithm must not average signals that don’t belong together, so each pack speaks to one person.
If the ad doesn’t call them out, it won’t reach them.
Cold or ready to buy, before production starts. An ad without a layer isn’t finished.
Each pack gets a full week and its share of the budget before we judge it.
Raw ROAS makes retargeting look better than it is, because it takes credit for people who were going to buy anyway.
We know which ad or which piece of research it came from, or it can’t be judged.
Reading the results
We stop at the first bad rung, because nothing below it can be judged yet.
Judged after about 100 clicks. No single good number exists, it depends on the industry. If it is off, we check placements and whether the ad meets policy.
Judged after the first few thousand impressions. If it is low, we change the hook and the first frame and leave the body alone.
Compared with your own winners. If it drops, we rewrite the body, the pacing and the proof.
Always read by layer: lower for cold traffic, higher for buyers. At half the cost per click, the cost per purchase often turns around.
We want 85 to 90%. Less means a slow site or bad placements.
Only after spending 1.5 to 2 times the break-even cost per purchase, and always against break-even. We never scale before this point.
The rest of the system
Each one runs alone, or all of them run together in the marketing package.