A/B test creators in cohorts, not one-offs

Randomly trying creators teaches you nothing transferable. Run them in cohorts with controlled variables so each test produces a reusable rule.

Cohort setup (owner: you, trigger: monthly):
— Group 5-8 creators with similar audience size + niche
— Hold the offer, creative angle, and landing page constant
— Vary one thing: format (reel vs video), or hook style, or CTA placement

Measure on RPC, not raw revenue:
— Bigger creators will out-earn on volume; RPC tells you who's actually efficient

If X then Y:
— If a format wins by >20% RPC across the cohort → make it the default in next briefs
— If results are within noise → keep both, decide on availability

Definition of done: cohort has a documented variable, fixed controls, RPC table, and one written learning added to your playbook.

Rule: change one variable per cohort. Two variables and you can't attribute the win. Every cohort should end with a sentence you can paste into a future brief — that's the compounding asset.