In this article, you’ll learn how to conduct an experiment that compares the performance of 2 different sets of assets (an "A" set vs. a "B" set) within the same asset group to find out which creative combination works best for your Performance Max campaign.
On this page
- How it works
- Set up an experiment to compare different sets of assets
- Keep in mind
- How to action on your A/B test
- Fix issues creating an experiment
How it works
When you set up an asset comparison test, your assets are divided into 3 categories:
- Control group (assets A): A selection of your existing assets that you want to use as a reference for comparison.
- Treatment group (assets B): A selection of assets that you want to compare against the control group. This can include other existing assets or new assets you upload for the test.
- Common assets: Any assets in the asset group that you don't select for either the control or treatment group. These assets will continue to serve normally to 100% of your campaign's traffic, alongside the control and treatment assets in their respective traffic splits.
Set up an experiment to compare different sets of assets
- Go to Experiments within the Campaigns menu
.
- Select the plus button
under the “All experiments” tab.
- Under “What do you want to test?”, select Assets.
- Select Assets provided by you under “Choose a variable to test”.
- Select “Performance Max” under “Campaign type” and select Continue.
- Select Any assets under “Select experiment type”.
- Select Select campaign to choose an existing Performance Max campaign and the Asset group you want to test.
- In the “Control arm” card, you can view your existing assets.
- In the “Treatment arm” card, select or upload the assets you want to test and compare against the assets in the control group. You can choose to add additional assets or completely switch to new assets in the treatment arm.
- Define the traffic split percentage for the "Control" and "Treatment" arms.
- Rename your experiment by selecting the “Experiment name” field.
- The experiment's start date is automatically set to the following day.
- The end date is determined by the Experiment Guidance System, which calculates the necessary duration for statistically significant results based on the current campaign selection. While you have the option to modify the end date, it is recommended to run the experiment for a minimum of 4 to 6 weeks to ensure best practices are followed.
- Select Schedule.
Keep in mind
- Locked assets: When an experiment starts, you can’t edit, add, or remove any assets in the asset group that’s being tested until the experiment ends. The asset editing pages for that asset group will be in a "view-only" mode to ensure the validity of your test results.
- Asset approval: Newly uploaded assets for the treatment group go through the standard policy review process. Disapproved assets will not be eligible to serve in the experiment.
- Asset Limits: Both the control and treatment assets will count towards the asset group limits.
- Asset Group Limitation: You can only test assets within a single asset group per experiment.
How to action on your A/B test
When you select Apply experiment, it updates your original asset group with the assets from the treatment group. You will have 2 options:
- Add treatment assets to campaign: This adds the "B" set of assets (the assets you want to test against the control assets) to your asset group.
- Keep control assets in campaign: This option is selected by default. It means your "A" set of assets will also remain in the asset group. If you uncheck this box, the "A" set of assets will be removed from the asset group.
If you select End experiment, it stops the test without making any changes. Any new assets from the treatment group are discarded, and your asset group reverts to its original state.
Fix issues creating an experiment
In some instances, you may encounter an error when trying to create an experiment. Here are a few reasons why this could happen, along with steps you can take to fix these errors.
- If your original campaign has been removed, try creating an experiment from a different campaign.
- If your original campaign already has a shared budget, is part of a portfolio bidding strategy or is using Smart Bidding Exploration, remove your campaign’s shared budget.
- If your original campaign already has an experiment running or is scheduled during your selected dates, try selecting dates that don’t coincide with another experiment.
- If your campaign has features that aren’t supported by experiments, try selecting a different campaign to convert into an experiment.
- If keywords with policy issues were excluded and haven’t been copied to your experiment, you’ll need to find and edit your keywords before those keywords can be copied to your experiment.
- If your asset group has more than 15 assets, it can be maxed. You can try removing some assets to bring them within the limit.