In the Advantage+ era, audience targeting is largely automated. The variable you control is creative. The brands winning on Meta in 2026 are the ones testing 15-20+ creative variants per month and systematically identifying what resonates. Here is our framework.
The Testing Structure
We run a dedicated creative testing campaign alongside our main ASC campaigns. Budget: 15-20% of total Meta spend. Setup: broad targeting (no audience restrictions), cost cap bidding at 1.5x your target CPA (to give creatives room to get data), and each ad in its own ad set (for clean data separation).
Creative Categories to Test
Every testing cycle should include variants across these categories: static product images (studio shot, lifestyle, flat lay), UGC-style content (testimonial overlays, unboxing, before/after), short-form video (under 15 seconds, hook in first 3 seconds), carousel formats (product features, use cases, social proof), and text variations (different headlines, body copy, CTAs) on the same visual.
The Decision Framework
After 3-5 days (or after each creative has 500+ impressions and $20+ spend), evaluate. Primary metric: cost per purchase (not CTR, not engagement rate — purchases). Secondary metric: hook rate for video (percentage who watched past 3 seconds). Kill anything with CPA above 2x target. Promote anything with CPA below target to the main ASC campaign. Everything in between gets another 3-5 days of data.
Scaling Winners
When you find a winning creative (CPA below target, 10+ purchases), add it to your main ASC campaign immediately. Do not increase the testing campaign budget — instead, let ASC spend more on the proven creative. We typically find that 1 in 5 tested creatives becomes a winner, and the top performer often delivers 3-5x better CPA than the worst performer. This is why volume matters.
Creative Fatigue
Every winning creative eventually fatigues. On Meta, this typically happens after 2-4 weeks of heavy spend. Watch for: rising CPM, declining CTR, and increasing CPA. When fatigue signals appear, refresh with new variants using the same concept (same hook/angle, different execution). Do not kill the concept — iterate on it.