How to A/B Test Ad Placements with Testing Groups

Testing Groups let you run controlled A/B tests on your ad placements. You split your traffic into groups, serve a different placement configuration to each group, and compare the results in reports to decide which configuration performs better.

Typical use cases include testing ad refresh settings, ad sizes, insertion types, lazyload thresholds, bidder line-ups, or floor price strategies — anything that can be configured on a placement.

How It Works

• Your traffic is divided into 100 equal buckets. Each visitor is randomly and consistently assigned to a bucket.
• A test contains two or more groups. Each group receives a traffic share (in percent) and is assigned that number of buckets.
• Groups of the same test never overlap: a visitor belongs to at most one group per test, so the comparison between groups is clean.
• Different tests run independently at the same time. You can, for example, run a refresh-rate test and a floor-price test concurrently without them interfering with each other.
• Bucket assignments are stable. Saving a test again does not reshuffle the buckets of existing groups, so a running test is never disrupted.

1. Create a Test

From the sidebar, navigate to Features > Testing Groups, then click New Test (top-right).

Use the + button to add more groups. A simple 50/50 test has two groups, for example "Control" at 50% and "Variant" at 50%. You can also test on a subset of traffic, for example 10% / 10%, leaving the remaining 80% of traffic outside the test.

Click Save Changes. The platform randomly assigns traffic buckets to each group according to its traffic share.

2. Start an A/B Test on a Placement

From the sidebar, navigate to Projects > All Projects, select the project, then click Placements.

In the placement list, click A/B Test on the placement you want to test. The "Duplicate for A/B Testing" dialog opens:

1. Select the test to use. All groups of one placement's test set must belong to the same test, so once a variant exists the test is locked.
2. Pick a testing group for the existing placement (for example, Control).
3. Pick a testing group for the duplicated placement (for example, Variant). The two placements must use different groups, and a group already used by another variant of the same placement cannot be selected again.
4. Click Duplicate.

At least two unused enabled testing groups are required in the selected test. If none are available, define more under Features > Testing Groups first.

The result:

• The original placement now serves only the traffic of its assigned group.
• An exact copy of the placement is created, assigned to the other group, and shown nested under the original placement in the list.
• The copy is created disabled so you can adjust it before it goes live.

Open the copy's Settings, change the configuration you want to test (for example a different refresh rate or insertion type), then enable it. From this point, the two configurations serve to separate, non-overlapping slices of your traffic.

You can also assign a testing group to any placement manually: in the placement form under Pre-Targeting, set the Testing Group field. The default value ALL means the placement serves all traffic and is not part of any test.

3. Compare Results in Reports

The report pages include a Testing Group filter. Filter values are labelled as "Test / Group" (for example, "Refresh Rate Test / Variant") and support equals, does not equal, in the list, and does not in the list operators.

Use the filter to compare the key metrics between groups — impressions, win rate, average eCPM, and page RPM — and let the test run long enough to collect a statistically meaningful amount of data before drawing conclusions.

4. Conclude the Test

When a winner is clear:

1. Apply the winning configuration to the original placement.
2. Delete the losing variant placement. When the last variant of a placement is deleted, the testing group assignment is automatically cleared from the original placement, so it serves all traffic again.
3. Optionally delete the test under Features > Testing Groups. A test cannot be deleted while any of its groups is still assigned to a placement — remove it from the placements first.

By following these steps, you can safely experiment with placement configurations on live traffic, measure the real revenue impact, and roll out only the changes that are proven to perform.