I've worked at Meta and what I've learned there is that the only real testing will happen with the AB testing. Everything else we do (CBO) is something that media buyers interpret as "testing", while it's, in fact, giving the algorithm open hands to do as it pleases. So, the difference between AB testing and a CBO is this:
- AB testing takes different audience segments from our audience set up and makes sure it doesn't overlap. When it shows the ads to the audience, the system makes sure it shows them to a different audience group (within our audience) and spends the same amount, basically prioritising the creatives and not the budget.
- CBO distributes the budget as the system sees fit at that period, and if one ad set has better performance than the other, CBO will push more budget there, therefore giving other ad sets less opportunities ----which is not real testing!
On top of it, the audience in this case overlaps and we can't actually attribute a purchase to one ad over another because we don't know if someone from that audience saw another ad before the winning one which made the person make the purchase.
And the learning phase - every time you add a new ad set, the campaign has to learn from a new angle and it brings it back to the learning phase, whether we see it or not - that is what is happening in the background.
WHICH CREATIVE WORKS?
None all the time. They constantly change because people become immune to whatever becomes popular.