<b>SOP: Launch a clean A/B test</b>

<b>SOP: Launch a clean A/B test</b>
Run this whenever someone proposes a test, before you spend a dollar on it.

<b>Stage 1 — Define</b>
— ☐ One variable changed, everything else identical (Owner: buyer)
— ☐ Hypothesis written as "changing X will move Y because Z" (Owner: buyer)
— ☐ Primary metric and minimum effect size agreed up front (Owner: buyer)

<b>Stage 2 — Size</b>
— ☐ Required sample/conversions calculated, not guessed (Owner: buyer)
— ☐ Stop date set so you don't peek and call it early (Owner: buyer)

<b>Stage 3 — Run</b>
— ☐ Budget split evenly, traffic source held constant (Owner: buyer)
— ☐ No mid-test edits to either variant (Owner: buyer)

<b>Stage 4 — Conclude</b>
— ☐ Result logged win, lose, or inconclusive with the numbers (Owner: buyer)

<b>Owner:</b> Media buyer
<b>Trigger:</b> Test proposed
<b>Done-when:</b> Sample size reached and a verdict is logged.

A test you stop early is an opinion with a budget. Save this. Run it every time.