Evaluate experiment results

Once an experiment has started, you can check its status and review its progress while it’s running. When an experiment has ended, the variation or control group with the higher revenue lift wins.

By default, auto-pause is now on for all new experiments. Auto-pause is a safeguard that limits cost and risk by checking in on your experiment results once a day and automatically pausing underperforming experiments based on conditions you set. You can turn off auto-pause for an experiment at any time.

Manual experiments and experiments started from opportunities can be evaluated using the same steps.

View related Skillshop training course

Running experiments now show near relevant entities to help you make data-driven decisions. To start a new experiment or review experiment results, look for the Experiments icon "".

Complete the following steps to review the results of an active or completed experiment:

1. Sign in to Google Ad Manager.
2. Click Optimization Experiments.
3. Review the experiment summary in the Experiments table. Only active, paused, or completed experiments are listed. The summary includes:
   • Name: The name of the experiment.
   • Experiment type: The format used for the experiment, if assigned.
   • Status: Indicates whether the experiment is actively running, paused, or has been completed.
   • Results: The outcome of the experiment. Active experiments may display the expected results if enough data is available.
   • The control is winning: The control group has experienced the highest revenue lift during the experiment period.
   • The experiment is winning: The variation group has experienced the highest revenue lift during the experiment period.
   • Insufficient data: Experiments require a significant amount of data before a recommendation is possible. Run your experiments for at least 7 days to improve the chance for conclusive results. Note that AdSense backfill data is not included in experiments.
   • Result is inconclusive: Experiment results may be inconclusive if there's no difference in revenue between the variation group over the control group or the improvement is statistically insignificant (<0.5% difference).
   • Experiment period: The dates the experiment will run. May be future dates for experiments that are scheduled, but haven’t started yet.
   • Daily revenue impact: The estimated change in daily average *revenue (based on data through the previous day) that would have been earned if these experiment settings were applied to 100% of network traffic. If you paused the experiment while it was running, this value may be lower than expected.
   • Experiment eCPM: The change in matched eCPM* (revenue∕matched queries) as a result of the experiment settings when compared to the control settings.
   • Experiment coverage: The amount of change in the number of ad requests that generated impressions as a result of the experiment settings when compared to the control settings. Measured in percentage points to represent this difference.
   For experiments, revenue refers only to non-line item and non- mediation demand.
4. Click the experiment in the table to view expanded details, including traffic allocation, pricing, trials, and any experiment types settings.
5. Decide how to proceed with the experiment. If you make no selection, the experiment continues until the scheduled completion date. You can also come back to view the results and make a decision at a later time.
   • Click Pause to stop an experiment from running. You can resume the experiment at any time.
   • Click Apply variation to end the experiment and implement the changes for all impression traffic immediately. 
   • Click Decline variation to stop the experiment and remove it. We'll ask for feedback to help us improve our future recommendations, and we won't suggest this particular experiment to you again.

Once you use, dismiss, or delete an experiment, it no longer appears on the Experiments page.

_{*Revenue does not include remnant line items.}