<b>Split test sign-off — when an A/B result is real</b>

<b>Split test sign-off — when an A/B result is real</b>

Calling a winner early is how you scale a loser. Use a fixed sign-off checklist before declaring a test done.

— Set the minimum conversion count per arm BEFORE launch — aim for at least 25-30 conversions per variant, not per click count
— Confirm both arms received roughly equal traffic (within 5%) — uneven splits invalidate the comparison
— Verify both arms ran across the same time-of-day and day-of-week window — Monday vs Saturday traffic differs
— Check the result holds across your main segments (device, top 3 GEOs) — a blended winner can hide a segment loser
— Confirm the lift exceeds your noise floor: a 3% difference on 200 clicks is noise, not signal
— Rule out an external cause: an offer cap, a creative disapproval, or a tracking gap mid-test

If the test fails any line, it isn't done — extend it or rerun clean.

Document the winning variant's ID and the sample size that decided it.

Save this SOP. Run this before every launch.