The direct answer
Test one variable — hook, visual format, or offer — against a clear control, run it long enough to reach a meaningful sample, and ship a new test every week rather than letting a single test run for a month. Weekly cadence catches winners and losers faster than it introduces noise.
What to test first
The hook — the first 1-3 seconds of a video or the headline of a static ad — has outsized influence on performance relative to the rest of the creative. It's almost always the highest-leverage variable to test before format or offer.
Avoiding false conclusions
A test that hasn't reached a meaningful spend or impression threshold will show noisy, misleading results. It's better to run fewer, cleaner tests with adequate sample size than many underpowered ones that produce contradictory 'winners' week to week.