Different ads. Too few different ideas.
The paid media team had increased creative output, but its test plan kept asking the same customer argument to perform in different wording. ‘Save time’, ‘work faster’ and ‘reduce manual effort’ represented one proposition rather than three independent hypotheses. More assets were consuming budget without broadening what the team could learn.
Choose the ideas worth spending on
Generating another route costs little compared with giving it enough exposure to reach a decision. The test plan therefore needed fewer simultaneous routes than the ideation process produced, with comparable audiences and a clear downstream measure.
A novel idea still had to connect to a documented objection and an approved claim. The gain from AI was breadth before selection, while the media budget remained concentrated enough to distinguish a useful argument from auction noise.
Test a different reason to buy.
Changing a headline or image is not always a new hypothesis. Each test needed a different customer concern, a supporting argument and a result that could change the next campaign.
- Customer barrier
- An objection supported by research or sales evidence, rather than an invented persona insight.
- Hypothesis
- A specific belief the message seeks to change and a qualified behavior expected to follow.
- Proof
- A claim the business can substantiate, not a persuasive detail supplied by the model.
- Decision
- What evidence would justify further development, rejection or another test.
AI widened the options. Customer evidence and live testing decided which options were worth keeping.
From buyer concerns to a usable test plan
Sales questions, lost opportunities and earlier tests supplied the concerns worth addressing. Each creative route named the concern, the promise and the proof, so a different visual did not automatically count as a different idea.
I selected tests the budget could support and recorded the decision afterward: develop the argument, reject it or investigate further. A rejected idea could still save the team from funding the wrong message again.
What the twelve tests showed
Over ten weeks, the team tested twelve distinct hypotheses rather than five, and three tests produced useful decisions. The strongest validated variant improved qualified conversion by 18% against its relevant baseline.
Twelve distinct hypotheses produced three useful decisions. The +18% qualified-conversion result belongs to the strongest validated variant, not to the entire creative program or every AI-generated ad. Follow-up exposure remains important as audiences and creative fatigue change.
A creative test earns its budget when the answer changes the next campaign. Three useful decisions mattered more than twelve launches; the winning variant was one finding, not a promise for every asset.