Creative testing frameworks for paid ads
A systematic way to test ad creative so you actually learn what works.
Most ad creative testing is random. A few variations get shipped, the one with the best CTR gets praised, nobody learns anything transferable. A real framework changes that.
The variables to test
Group your creative variables into three layers.
Concept. The big idea. The angle. "Founder talks about a problem" vs "Customer testimonial" vs "Product demo". Different concepts attract different audiences.
Execution. Within a concept, the specific creative. Different hooks, different settings, different actors, different voiceovers.
Format. Video vs static. 9:16 vs 1:1. With caption vs without.
Test one layer at a time. Mixing concept tests with execution tests means you can't isolate what mattered.
How to structure a concept test
Take 3-5 different concepts. Produce one strong execution of each. Run them all to the same audience with the same budget for 1-2 weeks. The winning concept is the one that beats the others at the metric you care about.
Now you have a hypothesis: "Founder-talking-head outperforms customer testimonial for our brand at this stage". You haven't learned this is true generally; you've learned it's true today.
How to structure an execution test
Pick the winning concept. Produce 5-10 different executions of it. Same idea, different angles. A/B test the executions head to head.
The winning execution might be 2-3x the loser. This is where ROAS gains actually come from.
How to structure a format test
Take the winning execution. Produce it in multiple formats: vertical video, square video, static, carousel. Some formats work better on some placements.
Many advertisers stop after the format test and never restart concept testing. Concepts age. Customer fatigue is real. Restart concept testing every quarter.
The Brian Balfour framework, adapted
A clean way to think about this:
1. Hypothesise an angle worth testing. 2. Produce minimum viable creative to test it. 3. Set a clear success threshold (e.g. CPA below $X for 14 days). 4. Run, decide, kill or scale. 5. If killed: learn why, adjust the hypothesis, retry. 6. If scaled: ride it until performance decays, then start over.
The mistake nobody admits
Most "creative testing" is actually creative production. The team produces a lot of ads, ships them, sees what survives. There is no hypothesis, no framework, no learning that transfers.
This is fine for some categories (high-volume, fast-fashion-style brands where creative is a constant stream anyway). It's wasteful for everyone else. A structured test cycle takes 2-4 weeks per layer but produces transferable learning that compounds.
AI's effect on this in 2026
AI generation has made creative production cheap. The bottleneck has moved from "can we make 20 ads" to "do we know what to make 20 ads about". Testing frameworks matter more than ever because volume without strategy just wastes money faster.