You can now A/B test two asset sets inside one PMax asset group.

This is a beta experiment. Here is how it works:
Control group: your existing assets
Treatment group: the alternative assets, either new creatives or ones you already use
Traffic split: you decide how traffic is divided between the two

You pick a success metric, such as conversions, value or clicks.

Google's Experiment Guidance estimates how long the test needs to run to reach statistically significant results. Google recommends at least 4 to 6 weeks.

Spotted by Thomas Eccel.