Experiments FAQs

Running several experiments at once isn’t recommended. They can interfere with each other, which can influence the result, meaning the results are less reliable. Instead, run experiments one after another. This sequential plan gives you clear data for every test.

After your experiment reaches the end date, Google will identify whether your experiment results were favorable. If your results are favorable, the experiment changes will be applied to the control campaign.

This happens often when you have deprecated or ineligible ad types, you should remove all the ads with deprecated ad types before restarting the creation.

Moreover, check if you have similar audience lists because campaigns with similar audiences, including those with removed similar audiences aren’t supported. Learn more About optimized targeting.

By default, this feature will be enabled when creating a new experiment, but you can choose to opt out before the experiment ends.

No, for all experiment types, including custom experiments and Demand Gen experiments, the traffic split percentage cannot be changed after the experiment has been set up. If you want to run an experiment with a different traffic split, you'll need to create a new experiment.

No, we don't apply any experiments which were ended manually. We only apply experiments which end on the end date that you define.

• If you’re using Max conversions with target cost per action (CPA) bidding, your experiment will be directly applied if conversions in your treatment arm are higher than your control arm, with CPA being lower.
• If you’re using Max conversion value with target return on ad spend (ROAS) bidding, your experiment will be directly applied if the conversion value in your treatment arm is higher than your control arm, with ROAS being higher.
• If you’re using Max conversions or Max conversion value bidding, your experiment will be directly applied if either the conversions or conversion value in your treatment arm are higher than your control arm.

It’s recommended that you run the experiment for at least 4-6 weeks or longer if you have a long conversion delay. We recommend you wait for 1-2 conversion cycles. Learn how long it takes for your customers to convert.

To maximize experiment power, pick campaigns with high volumes and run experiments for longer. We recommend running the experiment for at least 4-6 weeks. Keep in mind that if the experiment has many comparable campaigns with a significantly higher budget relative to the Performance Max campaign, just the noise from the experiment may overshadow the uplift from running Performance Max. Choose comparable campaigns thoughtfully in order to get detectable results.

It’s recommended for experiments to run for at least 4-6 weeks, and the first 7 days of data is discarded to account for the experiment ramp-up time. This ensures that you’re evaluating both arms fairly. For example, if your experiment start data is December 1 and the end date is December 31, you’ll only view data for the period between December 8-31 in the Experiment results page. However, you should be able to view stats for all campaigns in the main campaigns table for your desired date range.

For certain Search, App, and Display experiments, Google supports budget splits, while for some others, like Performance Max, Demand Gen, and Video, Google supports a traffic split.

• When you choose a budget split, Google divides the total budget across the arms, for example if you choose a 70-30 split, then the 70% budget will be allocated to control and 30% to treatment.
• When you choose a traffic split, Google divides the traffic, such as how many times the ads enter auction, across the arms, for example if you choose a 70-30 traffic split, then 70% traffic will go to control and 30% to treatment.

If the experiment type you're testing only supports traffic split, then it may or may not affect the budgets for your experiment. Roughly, when you test a change and it helps unlock additional value, you may spend more. Make sure that the campaigns aren't budget constrained, and the campaign will spend its full budget at maximum.

Google respects responsible spending guidelines, so while your experiment campaigns can spend up to twice your daily budget on a given day, you won’t exceed 30.4 times your daily budget over the month. To check if you've been credited for overdelivery, compare “Served Cost” vs. “Billed Cost” in the Reports Editor. Avoid frequently pausing your campaigns manually to manage spend, as this can disrupt the machine learning process. Learn Why daily costs might exceed your average daily budget.

If you choose a traffic split that isn’t 50/50, Google scales the results to make reporting easier to understand. For example, Google might scale the results of each arm up to 70% or down to 30%, even though the actual traffic distribution remains at 70/30.

Comparable campaigns are campaigns that are similar to the Performance Max campaign and may serve on the same inventory as Performance Max campaigns. They’re included in the control and trial groups of your experiment.

Comparable campaigns are automatically selected for your experiment based on factors such as:

• Matching domain names
• At least one overlapping conversion goal
• Overlapping locations

This is necessary to ensure the correct experiment setup.

Yes, you can edit comparable campaigns. Below you’ll find details to learn more about reviewing and editing your campaigns.

Performance for existing campaigns won’t be negatively impacted by experiments. Putting them into base and trial simply tags which ones had Performance Max traffic alongside them and which ones didn’t. For the existing campaign traffic in the trial arm, Google will measure what happens when it runs alongside Performance Max. This should capture any effects of traffic shifting as well.

Make sure to target the same products in both the Standard Shopping campaign and the Performance Max campaign in order to accurately test the performance of both campaigns. Ensure that the products targeted in the 2 campaigns aren't targeted by any other campaigns outside of the experiment. This will help ensure that the experiment won't interfere with existing campaigns in the account.

Traffic split options are available for Shopping versus Performance Max campaigns, but unavailable for non-GMC (Google Merchant Center) Uplift Experiments. When an experiment is scheduled, the traffic split can't be changed.

Yes, it’s technically possible to run other types of experiments in the same account. However, it's recommended you minimize these if possible.

Yes, you can. However, it’s generally recommended to make as few changes as possible while an experiment is in progress.

Yes, you can make changes to campaigns, then Google will automatically pick them for comparable campaigns and apply them to base and trial. However, making changes while an experiment runs isn't recommended.

This can be expected if changes are made to the campaign that are around conversion goal, domain, or location. If campaigns are added or removed, or the above changes are made in existing campaigns, additional comparable campaigns can be added or removed.

Existing campaign settings aren’t affected by being in the experiment. Keep in mind:

• Any existing Performance Max campaign that’s part of an experiment may notice a decrease in traffic because the campaign will only serve to 50% of the eligible traffic. When the experiment ends, Performance Max traffic should recover to pre-experiment levels if you launch the performance Max campaign.
• Any new Performance Max campaign created as part of this experiment will notice an increase in traffic if it’s launched to 100%.

The higher your Performance Max budget and spend is compared to the total spend in your account, the higher your chances are of noticing statistically significant results.

Set a budget you’re comfortable spending for the experiment despite the traffic suppression. Even if ads serve only 50% of the eligible traffic, you might end up using the entire budget. Remember, your budget can be doubled (the daily limit) similar to standalone campaigns.