A concept is an angle, not an ad. "Morning routine" is a concept. "Sarah's 30-second morning routine with the product on her bathroom shelf" is an ad. The difference matters because you're trying to learn which angles your audience responds to, not which exact frame went viral.
Get this distinction wrong and you'll spend a quarter chasing executions instead of building a creative library.
The brief that makes testing possible
Every concept brief should answer four things:
- What angle are we testing? One sentence.
- Who is this for? Specific. "Women 28-40 who follow Glossier" is more useful than "skincare buyers".
- What's the implicit promise? What does the viewer expect to gain or feel?
- What does success look like? Hook rate above X. CTR above Y. CPA at or below Z.
If a brief skips step 4, you have no shared definition of winning. You'll argue about results forever.
How many concepts to run at once
For most accounts spending under £30k/mo on Meta, 2-3 fresh concepts per week is the right cadence. Each concept ships with 3-4 variants. That's 6-12 new ads weekly.
Why not more? Because each concept needs enough budget to learn from. Splitting £5k across 20 ads gives you 20 inconclusive results. Splitting it across 8 gives you actual signal.
Structure: separate testing from scaling
Run a dedicated "testing" campaign with its own budget. Don't drop new concepts into your scale campaign and let the algorithm starve them. Testing campaigns get even distribution; scale campaigns get the proven winners.
- Testing campaign: ABO, even budget per ad set, one concept per ad set, 3-4 variants per ad set.
- Scale campaign: CBO or Advantage+, only ads that beat the threshold in testing.
The friction of moving an ad from test to scale is a feature. It forces a conscious decision: this won, here's the next step.
When to call a winner
The temptation is to call winners after 24 hours. The reality is most concepts need 3-4 days at meaningful spend to show their real CPA. Day-one performance is often misleading because of who Meta serves first.
Wait for:
- At least 3x target CPA in spend
- At least 100 link clicks (preferably more)
- Stable performance for 48 hours, not a single good day
If all three are true and CPA is at or below target, promote. If two are true, give it another 48 hours. If one is true, kill it.
The losers tell you more than the winners
Most teams celebrate winners and quietly delete the losers. That's backwards. A losing concept tells you what doesn't land with your audience right now. That's directional intelligence for next week's brief.
Keep a simple "concept post-mortem" doc. For every losing concept, note: the angle, the hypothesis, the actual result, and your best guess at why. Patterns emerge after 8-10 entries. You'll start seeing what your audience reliably rejects.
Common mistakes
- Testing creative changes inside the scale campaign. The algorithm protects spenders and starves newcomers. New creative needs its own space to learn.
- Killing a concept because the first ad in the set looked bad. Read the concept-level numbers, not the single-ad numbers.
- Re-testing a losing concept with the same hook six weeks later because "the audience might have changed". They probably haven't.
- Briefing concepts your team can't actually execute. A "premium cinematic" concept brief means nothing if you only have iPhone UGC footage. Brief what's shootable.
What good testing produces over time
After 8-12 weeks of disciplined concept testing you should have:
- A shortlist of 4-6 angles that consistently work
- A blacklist of angles that consistently flop
- A creative brief template that any creator can pick up
- A scale campaign that runs proven winners with confidence
That's the actual point. The library is the deliverable; the tests are how you build it.