Split testing (A/B testing)

분할 테스트 (A/B 테스트)

Shows ads that differ in one variable to mutually exclusive audiences so the comparison is fair.

Also known as: SPLIT_TESTSPLIT_TEST_V2creative testA/B test
···

A split test is like opening two shops in one city and arranging the shelves differently. If shoppers wander between them, the results blur, so the point is that each person sees only one side.

The developer docs describe it as comparing strategies on mutually exclusive audiences, with the API automating audience division and preventing overlap. You can test audience types, delivery optimization, placements, creative, budgets and more. Define KPIs and the confidence level first, and change only one variable at a time so you can find the cause. If A uses conversion optimization with automatic placements while B uses link-click optimization with custom placements, you cannot tell which change drove the gap.

Cells are split by percentage (treatment_percentage totals 100) and assigned at ad set or campaign level. A creative-only test uses type SPLIT_TEST_V2 with 2 to 5 cells, exactly one ad per cell, and creative_test_config holding either a daily budget or a lifetime budget percentage. The limits are 100 concurrent studies per advertiser, 150 cells per study and 100 ad entities per cell.

Evaluate with the efficiency metric for your objective; for conversions, the lower CPA wins. Do not compare groups of very different size or audience scale; scale them to be comparable first. Agree on the attribution model before starting.

When to use

Use it to verify whether a change to creative, optimization or placement really helps; one variable, comparable scale.

Open as page ↗

Suggest an entry

Tell us about a term, effect or component that’s missing. We read every suggestion.