A custom experiment makes a trial copy of a campaign before you change the real one, like test-selling a new menu item in one store before rolling it out. Apply a change, such as a Target ROAS bid change, to a trial campaign for Search, Display, Demand Gen or Video and compare it with the original. You choose the duration and how much of the original's traffic and budget it uses. For Search ad creative alone, ad variations fit better; for specific ad groups, use custom experiments.
The flow has three stages: set up, monitor against the original, then apply the experiment to the original (manually or automatically) or convert it into a new campaign. You can pick up to two primary success metrics.
Set expectations. A 50/50 split divides auction eligibility, not guaranteed impressions or spend, which depend on auction factors such as Ad Rank and bidding strategy. Allow 7-14 days for the treatment arm to stabilize. The campaign must be active, use an individual daily budget and a standard bid strategy, so shared budgets and portfolio bidding are out. Some reports, including auction insights and search terms, are unsupported. For reliable results the base campaign needs enough conversions; the page gives more than 100 daily conversions as an example. For general test design see the A/B testing and one-click ad experiments entries.
When to use
Use it before applying changes that are hard to undo, such as a bid strategy switch or targeting change.