Blog · Creative testing
Creative testing on Meta: test ideas, not outfits
Most creative tests on Meta compare versions of the same idea: two headlines, three backgrounds, a different crop. The winner tells you which outfit worked. It rarely tells you what to make next. Here is how we set up tests that answer the bigger question: which ideas are worth building on.
Every test has one of two jobs
- Explore: find out whether a new idea works. The idea changes; everything else stays as comparable as you can make it.
- Exploit: get more out of an idea that already works. The idea stays fixed; the hook, execution and format change.
Mixing the two is the most common reason test results can't be read. If the claim, the visual and the format all change at once, a winner tells you nothing you can reuse.
What counts as a new idea
An idea is a proposition: what the ad claims, and who it's for. A different reason to buy (price versus convenience versus peace of mind), a different buyer, or a different kind of argument (a demonstration versus a comparison versus a testimonial) is a new idea. A new colour, crop, presenter or setting for the same claim is not. It's the same idea in a different outfit. The briefing guide has the full rules.
How to set up an idea test
- Write each idea's angle in eight words or fewer. If you can't, the ad is arguing two things.
- Make the ideas distant. Each should differ in reason to buy, buyer or argument. Close cousins produce close results and no lesson.
- Keep the execution comparable. Similar format and production level across ideas, so the idea is what varies.
- Give each idea more than one ad. One ad per idea tests that ad, not the idea. Two or three executions per idea make the result about the claim.
- Use a clean split when the answer matters. Ads Manager's A/B testing splits audiences so ideas don't compete for the same people. Budget and time need to be enough for a result you'd trust.
Read results by idea, not by ad
Tag every ad with the idea it carries, then add up results per idea. An idea that wins across its executions is worth building on. An idea with one lucky ad usually isn't. Only then move that idea into exploit mode: keep the claim, and vary everything around it.
The benchmarks we use
These are Kaliber's starting points, not Meta rules. We adjust them per account as results come in. Kaliber benchmark
- With no results yet, the whole batch explores.
- Once some ideas are proven, start around 60% explore, 40% exploit.
- No untested idea takes more than 20% of the exploration share, so the new ideas aren't one idea in disguise.
- Offer-led ads stay a minority unless the offer is the strategy. Otherwise the brand teaches buyers that it is a price.
Know which ideas you're already testing
Before you brief a test, see how many distinct ideas your account runs today and how much each one carries. The Andromeda Autopsy reads your live Ad Library. Free, no login.
Kaliber is not affiliated with Meta.