My first AI ad batch for one brand lost to the ads already running in that account. Then I read the account properly. The offer people were buying on lived only in image ads, and nobody had made a video around it. The image ads had also dropped a quick at-home check for the problem. The next batch made both: a plain offer video and a talking head doing the check. Those two carried the batch.
Now I read all of this before batch one.
- Every format, not just video. Images, carousels and videos from the last 6 to 12 months, sorted by spend, not by ROAS. Spend shows what Meta chose to scale.
- The offer people actually buy on. The bundle, the price line, the guarantee in the top-spending ads. If it lives in only one format, the other formats are missing it.
- What got dropped. Lines, demos and checks that were in older winners and quietly disappeared. Those are often the first ads to make.
- The longest-running ads. What your buyer already answers to, and the month it stopped working.
- The comments on the top ads. The buyer's own words, and the questions they keep asking.
- Where the ads send people. Whether the first screen of the page says what the ad promised.
After batch one, I write down one thing I'm changing for batch two, and why. I'd ask any partner the same thing.
Ask after batch one
What did you change for batch two, and why?
If they can't name one thing, the next batch will look like the first.
