Custom experiments let you propose and test changes to your Search, Display, Demand Gen, and Video campaigns. You can measure your results and understand the impact of your changes before you apply them to a campaign.
This article explains how custom experiments work. When you’re ready, learn more about how to set up a custom experiment.
On this page
- Benefits
- How custom experiments work
- Traffic splitting and exposure
- Features that aren’t supported by custom experiments
- Keep in mind
Benefits
Simplified workflow: You can select your experiment type and create custom experiments in fewer steps. You can choose:
- Ad variations: Run text ad or responsive search ads on your campaigns or across your account
- Custom: Create a Search, Display, Demand Gen, or Video experiment
- Select the original campaign, modify test variables in the trial campaign, and run the experiment.
- Video: Test your video assets in your Video campaigns
Easier reporting: You can report across all your experiments from one place in the Experiments page. You can:
- View all experiments across ad variations in a single table, including Search, Display, Demand Gen, and Video experiments.
- View all other metrics in a summary table that compares performance for the original and experiment campaign (no hovering to compare stats).
- Identify up to 2 primary success metrics and get tailored reporting.
- View clearer performance comparison dates, with a date picker specific to custom experiments.
Manage experiments: Navigate through campaign management tables for original and trial campaigns.
- Navigate data across campaign, ad group, or creative levels (including regular columns like change history, among others) to help you thoroughly examine performance variations.
- Edit or update experiment variables from the campaign management table.
If you want to test your Search ad creative, we recommend using ad variations. However, if you want to test specific ad groups rather than the whole campaign, use custom experiments. To test video creatives, learn more about how to create a video experiment.
Example
Anthony and his boss consider changing bids for their campaign but want to be confident these changes will improve performance. Anthony creates a new experiment with Target ROAS bid changes and runs it for a month. Results at the end of the month show that changing bids improves performance, and the system automatically applies the experiment to the original campaign.
How custom experiments work
Custom experiments let you create custom experiment campaigns to test how your changes perform against your original campaign.
- Set up a custom experiment. When you set up your experiment, you can specify how long you’d like it to run and how much of your original campaign’s traffic (and budget) you’d like it to use.
- Monitor your custom experiments. As your experiment runs, you can monitor and compare its performance against your original campaign. If you’d like, you can change the dates of your experiment to end it early.
- Delivery expectations: A 50/50 traffic split controls auction eligibility but doesn't guarantee equal impressions or spend, which are determined by auction dynamics like Ad Rank and bidding strategy. Allow 7-14 days for the treatment arm to stabilize.
- Troubleshoot serving: If an experiment has stopped serving or has zero impressions, check if it's in a "Bid Strategy Learning" phase (typically 7 days). Don't make changes during this period and review for ad rejections, low Ad Strength, or conflicts between targeting and content ratings. Ensure that Consent Mode is active for EEA traffic.
- Apply your custom experiment. If your experiment performs better than your original campaign, you can apply your experiment to the original campaign (manually or automatically). You also have the option to convert your experiment into a new campaign with the same dates and budget as your original campaign and pause your original campaign.
Traffic splitting and exposure
When you set up an experiment, you choose how to split traffic and budget between your original and trial campaigns. This helps you compare results before you make permanent changes.
Even with a set split, one side might get more exposure. This happens because each ad slot is a new auction. If the experiment has different bids or assets, it may win more or fewer auctions. Changes to keywords or ad quality also affect how often an ad shows. Additionally, if one side hits its daily budget limit first, it stops showing ads for the day. Automated bidding may also shift volume while it learns which version works best.
Features that aren’t supported by custom experiments
Custom experiments generally support the same features as campaigns, with these exceptions:
- Ad customizers that use "Target campaign" or "Target ad group." However, you can create custom experiments and assign a target campaign or ad group feed after you create it. Learn how to Create ad customizers for responsive search ads.
- Ad schedule reports
- Auction insights
- Bid landscapes
- Category and search terms
- Display placements reports
- Keyword diagnosis
- Scheduled email reports
- Shared budgets
- Pausing or removing automatically created headlines or descriptions from the trial campaign
Keep in mind
- Custom experiments are available for Search, Display, Demand Gen, and Video campaigns. However, some limitations apply:
- Campaign eligibility: To select a campaign for an experiment, it must be active (not paused), use an individual daily budget (not a Shared Budget), and use a standard bidding strategy (not Portfolio Bidding). For Performance Max, check for conflicting brand guidelines or exclusions. If your campaign is missing from the selection menu, verify that it meets these criteria and allow a few hours for the system to sync.
- Performance Max: Use "Uplift" experiments to test alternate URLs. Note that "Experiment sync" may revert changes in your treatment arm to match the base campaign. Turn off sync if you need the treatment arm to maintain distinct settings.
- Video campaigns: Only auction-based campaigns are supported. Reservation-based products like YouTube Select or Masthead are incompatible.
- App and Shopping: Standard custom experiments aren't supported. Use "App asset uplift experiments" for App campaigns. Learn how to Create a new App asset experiment.
- Conversion volume: For reliable results, ensure your base campaign meets minimum requirements, such as more than 100 daily conversions.
- You might not view a campaign as an option if it uses a shared budget. AI Max experiments don't support shared budgets yet. To fix this, change to an individual campaign budget.
