In ResultsAiX Optimize a variant is called a version. Every test starts with two: "Original", which is the page as visitors see it today, and "Version B". Each starts with a traffic share of 50.
Adding and removing versions:
Press "Add a version" for "Version C", "Version D" and so on. There is no set limit in the builder. When you add one, the shares are rebalanced evenly, and you can then change them. Rename a version by typing over its name. The bin icon removes a challenger, and the confirmation tells you its changes and, once saved, its results are removed. The Original cannot be removed.
Traffic share:
Each version has a "Traffic share" number from 0 to 100. The shares work as weights, and the builder shows the result beside each one as a percentage of visitors. 50 and 50 is an even split. 70 and 30 sends about seven visitors in ten to the first version.
A visitor keeps their version:
The version a visitor is given is remembered in a first-party cookie for 180 days, so they see the same version on every return visit. With "Share of visitors in the test" below 100, a visitor who falls outside the test sees the original, is not counted, and stays outside on later visits.
Mode: Classic split or Autopilot:
The "Mode" box in section 4 has two choices.
- Classic split keeps the shares you set for the whole test. It gives the cleanest statistics and usually reaches a confident answer sooner.
- Autopilot gives each new visitor a version based on the results so far. A version that is currently doing better is more likely to be chosen, while the others still receive some visitors so the test keeps learning. When the winner is picked by revenue, Autopilot weighs order value as well. Your share numbers are not used in this mode. The trade off is a slower confident answer, because versions that are behind collect visitors more slowly.
Tests in Autopilot carry an "Autopilot" badge on the Dashboard, the Tests tab and the report.
Changing versions mid test:
Versions are locked while a test is running or paused. If you need to change one, pause the test, press "Reset results" on the report, edit, and start again, so the numbers describe one experiment.