One variable at a time
Golden rule of A/B: if 2 videos race, only ONE thing should differ. Different hook, same rest. Different scene, same hook. More than two variables: unreliable result.
4 variables worth testing
1. **Hook (first 3 seconds)** — highest impact. 2. **Scene** (café/office/outdoor) — medium impact. 3. **Duration** (7s/15s/30s) — low–medium. 4. **Aspect ratio** (9:16/1:1) — platform-dependent.
Budget formula
$5–10 per variant per day, 48 hours. With $20–40 total you learn which hook delivers the lowest CPC across 4 variants. That can mean 50%+ savings on the main campaign.
Stop criterion
Compare only variants that hit 1000 impressions. Below that, no statistical signal. Meta Ads is the cheapest learning platform for this budget.