A/B testing

An A/B test compares two versions of a design with real users. In the example, people are randomly split between a "Continue" and a "Review payment" button, and the team measures checkout completion.

Here the versions differ only in the button label, which isolates the effect of that change. A test can also compare a bundle of changes, but then its result applies to the bundle as a whole.

A/B tests show which version performs better, but not why. They also need enough traffic to give reliable results, so they suit live products more than early concepts.