Most creative tests on Meta don't test anything. Ten hooks on the same video, three button colors, two headlines: the system sees one creative, and you learn nothing you can reuse.
This method fixes the three things that matter: what you compare, how much you spend, and when you decide.
Test concepts, not variations
Meta groups ads that share visual and thematic attributes and treats them as variations of one creative. That has a direct consequence for testing: if your versions look alike, they compete for the same slot and you can't tell them apart. The background is in our Meta Andromeda guide.
A concept is a genuinely different angle: who's on screen, what they want, and how much they already know about the problem. A variation is everything else: hook wording, cut, captions, button. Test concepts first. Optimize variations once a concept has won.
How much budget, and per concept
Two numbers to keep apart.
| Rule | Source | What it means |
|---|---|---|
| 20 to 30% of total budget on testing | Meta, November 2024 | The share of spend that goes to new ads, with new ads every week |
| At least 2x target CPA per concept | Media buyer benchmark | The floor before judging a concept |
The second one isn't a Meta rule, and that's fine. It's simple arithmetic: below 2x your target CPA, a concept hasn't had enough budget to produce two sales at the price you want. Judging it earlier is judging noise.
If your target CPA is $40, each concept needs at least $80 before a verdict. With $2,400 of monthly testing budget, that's room for about 30 concepts, or 9 concepts carried by 3 faces each, with margin.
When to judge
- At least 7 days. That's the minimum Meta advises for an A/B test, which can run up to 30 days. A week also smooths out weekday effects.
- And the spend floor reached. Both conditions, not one.
- Longer if your buyers are slow. If customers usually take more than a week to purchase, extend the test.
- Don't touch the ad set. Budget changes, new ads and pauses are significant edits that can restart learning and blur the result.
Which tool to use
| Tool | How it works | When to use it |
|---|---|---|
| Ad creative testing (inside a campaign) | 2 to 7 copies of an ad in an existing campaign; Meta advises no more than 20% of the campaign's budget; no confidence level | Quick reads on concepts, in your real setup |
| A/B test | Each version goes to a separate share of the audience; nobody sees both | When you need a clean comparison between two strategies |
| Conversion Lift | A holdout group that doesn't see your ads | When you need proof of incremental effect |
The PDA matrix: briefing concepts that really differ
PDA is a grid to generate concepts on purpose instead of by accident.
- P — Person on screen. Who carries the message: age, life situation, setting, way of speaking.
- D — Desire. What they actually want: save time, feel good, look good, stop worrying.
- A — Awareness. How much they know: problem-aware, solution-aware, product-aware.
Eugene Schwartz describes five awareness levels. We drop unaware, which needs a story about a problem the person doesn't see yet, and most-aware, which is about the offer or retargeting. The other three test well as concepts.
Three values per axis give 27 cells. Pick 9 a month. Carry each with 3 faces. That's 27 creatives for 9 concepts, enough to learn which cell works and whether it's the idea or the person.
Example
A skincare brand, one month:
| Person | Desire | Awareness | Opening line |
|---|---|---|---|
| Nurse after night shifts | Look rested | Problem-aware | "Three night shifts in a row and nobody can tell." |
| New dad | Save time | Solution-aware | "I stopped doing a 6-step routine." |
| 50-year-old runner | Stop worrying | Product-aware | "I've tried every SPF. Here's why I stayed." |
Each row is a concept. Each one is then carried by three different people.
Reading results
- Judge the cell before the face. If all three faces of a concept fail, the concept failed. If one wins and two fail, it's the person.
- Look at cost per result first, then hook rate and hold rate to explain why.
- Keep a log. Cell, faces, spend, CPA, verdict. After three months it becomes your brief.
- Move winners into your main ad sets in batches, not one by one, to limit learning resets.
Partnership Ads fit this method naturally: each creator brings a different face and a different account. How to set them up is in our Partnership Ads guide.