Voice of customer
Surveys inside experiments
Tie a survey to an experiment and only its visitors are asked. The number tells you what happened, and the answers tell you why.
A survey can be tied to one of the website’s experiments, picked from a searchable list grouped by status. Only visitors enrolled in that experiment see it, and you can narrow it further to specific variants, so a question about a new checkout is only asked of the people who saw the new checkout.
An experiment tells you that variant B converted better. It cannot tell you why, and the why is what decides what you test next. Reading the same question per variant usually can: if the arm that converted better also scores markedly higher on effort, you have both the result and the mechanism behind it.
It works the other way too, and that is the case worth planning for. A variant that wins on conversion while scoring worse on satisfaction is a warning rather than a win, and it is completely invisible to an experiment reading conversion rate alone.
The builder is plain about the cases that would quietly ask nobody. Tie a survey to a draft experiment and it says no one will see it until the experiment starts; tie it to an ended one and it says no one will see it at all. Enrolment is the experiment’s own, so the survey asks exactly the visitors the experiment counts.