Bidding & Budgets

Experiments (Campaign Experiments and Drafts)

Google Ads experiments split a campaign's traffic between the original and a modified version so a change in bidding, keywords, ads, or settings can be measured against a control before being applied.

In practice

Changing a live campaign and watching what happens is not a test; seasonality, competitors, and randomness all move the numbers. An experiment holds those constant by running the original and the treatment at the same time on a randomized share of traffic. You set the split, the duration, and the metric, and Google reports the difference with confidence intervals.

Good experiment candidates are changes with a real hypothesis and enough volume to conclude: a bidding strategy switch, a target change, broad match with brand lists against phrase match, a new landing page. Poor candidates are low-volume campaigns and changes so small the test cannot detect them.

Experiments have quirks. While an experiment is in design, its settings are write-only and reads return nothing. Negative keywords added to one arm should be mirrored to the other so both stay comparable; bids and audiences should not be touched during the test. When it concludes, apply the winner or end it, then note the result.

RELATED TERMS

Written by Ilya Bulychev, Founder, Live PPC Ads. Part of the PPC Glossary.