Paid, rewritten: the targeting is automated, so the creative is the variable.
The platforms took the levers marketers used to pull. What is left to engineer is what the ad says.
What the old playbook assumed
That skill lived in audience building, bid strategy and channel arbitrage. Broad automated targeting has absorbed most of that, and the platforms optimise against their own objective whether or not it is yours.
Where the remaining leverage sits
In the creative and the offer: what the ad claims, what proof it carries, and whether it reads as yours or as the category average. When everyone generates ads with the same models, the ads converge, and converged ads compete only on price.
Why most paid testing teaches nothing
Variants that differ cosmetically cannot produce a finding, and B2B volumes rarely reach the point where a small difference can be read at all. Across 127,000 experiments Optimizely found only 12% won on the primary metric, and that is with consumer-scale traffic.
How we run it instead
Ads are generated as a set of genuinely different angles, checked for distinctiveness and for claims we cannot prove, cut by a simulated buyer panel that is only ever allowed to remove options, and launched against a spend limit and a stopping condition agreed beforehand. Nothing goes live and no budget changes without a person saying yes.
What carries forward
Not the winning ad — the trait that won: the hook, the proof, the angle. That updates a written belief and seeds the next round, which is how a paid programme gets cheaper over time instead of resetting every quarter.
Questions
Do you manage the ad accounts?
We run the creative, the structure and the measurement, and every launch and budget change is approved by you first.
What budget does this need?
Enough for a stopping decision to mean something. Below that we say so rather than spending it slowly and reporting impressions.
How does this relate to the AI answer work?
Paid buys attention now; the answer work earns selection later. They share the same claims and the same proof, which is why they sit in one system rather than two agencies.