← Playbooks
Creative testing framework
How we choose what to test, how big a test is, and when a result counts.
LiveOwner Sam Okafor· Updated Sep 12, 2026· 80% adoption
01Pick the variable
- Test one variable at a time: hook, talent, format, offer, CTA or length.
- Every hypothesis names the audience insight it comes from.
- Rank hypotheses by expected impact × ease. Media agrees the top three each month.
02Design the test
- Use the platform's native split or lift tool where one exists.
- Size for at least 90% confidence on the account's primary KPI, agreed with media before launch.
- Hold spend flat between arms. No mid-test edits.
03Read the result
- Won: lift on the primary KPI at 90%+ confidence. Lost: negative at 90%+. Anything else is inconclusive.
- Log every result, including losses, to the learnings library within 48 hours.
- A winner becomes the new control; the next test tries to beat it.
04Watch for fatigue
- Flag when the primary signal drops 15%+ over its window at frequency above 3.
- Refresh on the same concept first (new hooks), then rotate concepts.