Skip to main content
An A/B experiment divides traffic for one AdUnit into buckets so you can compare different Line Item configurations. This example uses Brand A’s Spring promotion Order on a shopping mall’s Promotion banner. Before you start, check the placement, allocation, Creative, and targeting setup.
The current experiment form lets you select an AdUnit. Placement is marked as coming soon and cannot be selected. The Experiments menu may not appear, depending on your deployment and access.

​
Prepare and create an experiment

1

Enter the basic information

Open Optimization → Experiments and select Add Experiment. In Basic information, enter a name, choose an unarchived AdUnit, and set the start and end times. The end time must be later than the start time.
2

Configure the buckets

In Buckets, configure the buckets to compare. Bucket names must be present and unique.
3

Review Line Items and save

Add the Line Items for each bucket, then review their Weight and Line Item Rotation. The first bucket is marked Control. Review the target, period, bucket proportions, and Line Items in the Setup Summary, then save. Advanced JSON settings are marked as development functionality.
Bucket proportions are relative integer weights, not percentages. For example, proportions 1 and 1 split traffic evenly; 1 and 2 assign a relatively larger share to the second bucket. A request needs a user ID to receive a bucket. With the same user and experiment IDs and an unchanged bucket configuration, the user stays in the same bucket. Changing the buckets or their proportions can change the assignment. The console shows proportions as normalized values. Each bucket can contain up to 10 Line Items. An experiment can use up to 15 bucket IDs; a used ID cannot be reused after its bucket is removed. Check that the selected Line Items are allocated to the chosen AdUnit. If you change the target, the first bucket may be refilled with allocations in Pending, Requested, or Published status, so check it before saving. The Line Item weight inside a bucket is a separate relative value used to choose among that bucket’s Line Items. Traffic proportion and Line Item weight serve different purposes. The available rotation methods are Sequential, Weighted, Random (default), and Score-Aware. Read the descriptions in the console, and keep the configurations comparable.
Experiments list

​
Start and end the experiment

New experiments are created as Scheduled. Review the configuration, then set the status to Running in the edit form and save. Traffic is affected only while the status is Running and the current time falls within the configured start and end period. Only one active experiment can target an AdUnit at a time. The status moves from Scheduled → Running → Completed. You can complete a running experiment, but you cannot reactivate a completed one. When you edit a running experiment, a warning says the change may affect statistics. Record what changed and when. Admins should also check the change history when reading results. After the configured end time, the experiment no longer affects traffic, but its status does not automatically change to Completed. To finish it, edit the experiment and save its status as Completed. Only scheduled experiments can be deleted.

​
Review and compare results

Open a running or completed experiment and review the Buckets and Bucket Analysis tabs. Admins also see the Change History tab. In Bucket Analysis, choose a metric, date range, and interval to compare bucket time series and table rows. The maximum range is 7 days for hourly, 60 days for daily, and 90 days for weekly or monthly data. If the range you select is too long, the console shortens it and shows a notice, so check the resulting start and end times. Hourly metrics are kept for 90 days; see the notice on screen. Analysis is unavailable for scheduled experiments. Before comparing results, check that the experiment was running during the selected period. Also check each bucket’s proportion, Line Items, and weights. Admins should also check the change history. Results show metrics for each bucket. They do not guarantee statistical significance or better performance, and a difference alone does not show what caused it.

​
Decide what to change next

If buckets differ, review impressions, clicks, conversions, and Creative-level metrics for the same period in Line Item Analytics. Check the AdUnit’s overall trend in Inventory Analytics as well. Determine whether the difference is limited to a time period or Creative, then consider changes to allocations, ad type, targeting conditions, or Creatives. Record what changed and when, then review a comparable period. If you need to separate the effect of a change from other conditions, design a separate experiment. The experiment feature does not optimize settings or pick a winner for you.