Generative tools have removed the bottleneck on ad creative. A team can produce dozens of headline, image and video variations in an afternoon. The new bottleneck is deciding what to test, because every variation you run costs money to reach a meaningful result.
Test hypotheses, not variations
Before generating anything, write down the questions you want answered: does a product close-up beat a lifestyle scene, does a price-led headline beat a benefit-led one, does a customer quote beat a brand statement. Then generate variations that isolate each question. Ten random creatives tell you little; three pairs that differ in one way each tell you a lot.
Use AI where it is strongest
- Drafting the long tail of formats once a winner is known: sizes, lengths, languages.
- Pre-scoring creatives for basic quality, readability and brand rules before spend.
- Summarising results across campaigns into plain-language findings.
Keep humans on the money decisions
Set a fixed test budget per hypothesis and a clear stopping rule. Let the platform’s own optimisation pick delivery, but let a person decide what the result means for the next round. AI is excellent at producing and summarising; it is not yet a substitute for judgement about what your brand should say.
