What's a best practice to follow when designing an experiment to test Performance Max campaigns?

The selection is correct because optimizing Performance Max to a comparable CPA or ROAS target to other campaigns is essential for conducting a "fair test" that accurately measures the uplift provided by AI-powered automation. When running an experiment, the primary goal is to determine if the new campaign type can achieve the same—or better—efficiency as your existing strategies; if the targets are mismatched, the AI's bidding behavior will be skewed, making it impossible to tell if performance changes are due to the campaign's effectiveness or simply the different target constraints. Setting consistent targets allows the machine learning algorithm to compete on a level playing field with your "standard" campaigns, providing a clear data-driven comparison of how Performance Max utilizes its cross-channel reach to find incremental conversions at your desired cost-per-acquisition or return on ad spend.