Glossary
What is A/B test (yield)?
Also called: A/B testing, holdout test, yield experiment
A yield A/B test splits visitors at random between your current ad setup and a change, then compares revenue to show whether the change really helped.
हिंदी में पढ़ेंBefore-and-after comparisons mix your change with the season, the news and traffic swings. In an A/B test, visitors are split at random, so both groups live through the same days. The group that keeps the current setup is the control, or holdout; the other group gets the change, such as a new bidder, floor or layout.
Decide the length before you start and run full weeks; stopping early when a result looks good picks a lucky moment. Google asks for Ad Manager manual experiment trials to run at least 7 days and up to 31. Judge the whole visit: compare revenue per 1,000 sessions, not CPM alone, and watch pages per session and page speed. An A/A test of two identical groups shows how big normal noise is.
For demand partners, keep a small holdout even after rollout. A new SSP’s dashboard shows what it won, not what it added; a holdout shows whether that money was new or just moved from another bidder.
Example
20% of visitors get a new bidder and 80% are the holdout. After two full weeks, the test group earns $10.30 per 1,000 sessions and the holdout $10.00. The lift is $0.30 ÷ $10.00 = 3%.
Why it matters to publishers
Without a proper split, you cannot separate a change from the calendar. A/B tests stop you keeping changes that do not work, or dropping ones that do.