Skip to main content
Consent records and statistics

End or troubleshoot a banner A/B test

End a test, keep version A or apply version B, and solve common problems.

1. Know the test states

A test is a Draft, Running or Ended. Discard draft deletes a test that has not started. A running test cannot be paused or resumed. Ended tests stay available as read-only history.

2. End the test

Select End test whenever you want, also when the result is still unclear after two days. In the confirmation, keep A, which is selected by default, or use B; you can preview both. Nothing is applied automatically.

Only the tested configuration on the selected domain changes. Other regional configurations and functional settings are kept, and applying B does not reset the consent visitors have already given.

To read the result before you decide, see Understand your banner A/B test results.

3. Check that publication succeeded

If you chose B, check the status after ending. If publication failed, B is not live: read the status or error message and retry as it describes. The results of the ended test stay read-only.

4. Solve common problems

Most problems have a simple cause:

  • The menu item is locked: check that the selected domain itself is on Business or Enterprise.
  • No banner can be selected: the domain needs a published opt-in configuration.
  • You cannot create a test: discard the existing draft or end the running test.
  • Start test is unavailable: B needs a permitted change from A.
  • The published banner changed after you created the draft: create a fresh draft.
  • You keep seeing A: a browser keeps its version during the test, so use the previews to inspect both versions.
  • Few choices appear: existing consent is not reset, only routed browsers that actually make a choice are counted, and previews add nothing.
  • The versions have different counts: a 50/50 split does not guarantee equal numbers of responses, and the Statistics page also counts later choices.
  • You need to change routing or consent settings: end the affected test first.
  • No clear difference: wide uncertainty is not an error, and you can still end the test.
  • Confidence interval unavailable: report it to support and do not choose a winner from the percentages.

5. Know what happens when the plan changes

If the domain leaves Business or Enterprise while a test is running, the test stops and visitors get A again. The results are kept, but A/B testing is locked until the domain is eligible again, and upgrading does not resume the old test. If a test had already ended with B applied, a later downgrade does not undo that published design. See Upgrade or downgrade your plan.

6. Test safely and contact support

Do not clear real visitors’ storage, submit choices repeatedly or make visitors choose again to improve the sample. For repeated checks, see Create repeatable consent test sessions. When you contact support, give the domain, the test name, the time with its time zone and the status shown. Do not send passwords or raw visitor records.