Maze — Examples

Example 1 — Is this a Maze question?

A team wants to know whether people can find the returns link, and separately whether a new “swap instead of refund” idea is worth building. Decide which question fits an unmoderated test, and rewrite the tasks that go with the one that does.

Question A: can first-time buyers find the returns link in the
            account area?
Question B: would people use a "swap instead of refund" option,
            and why or why not?
Draft tasks
  A1. "Please try to locate where you would go to return something."
  A2. "Imagine you wanted to swap an item. What do you think?"

Solution

Fit: A is countable (found the link / did not) and needs no
encouragement. B asks why, which an unmoderated tool cannot answer —
run two moderated sessions for B instead.

Rewritten tasks
  A1. "You want to return a pair of boots. Start the return."
  A2. "Find out how long you have to return an item."

Removed: the swap task, because it is a why question.

Maze suits a task that needs no encouragement and has an unambiguous metric, while the question about a new idea needs reasons the tool never collects. Plain, context-free instructions keep the rates from turning into noise. The usual failure is using unmoderated testing for a why question and then reporting paths as if they were reasons.

Example 2 — Reading the screen before the headline

A ten-participant run of the returns flow has finished. Read the per-screen drop-off rather than the completion rate, and state what the numbers cannot explain.

Maze report (illustrative figures)
  task completion            60%
  reached the account area   10 of 10
  reached order details       7 of 10
  reached the return reason   4 of 10
  reached confirmation        4 of 10

Solution

Finding: the return-reason screen loses three of the seven who arrive
there, and those three never continue. The 60% completion rate hides it.

What the data cannot say: whether those three misunderstood the wording
or found a genuine flaw. Book two moderated sessions on this step.

Report line: the sample was ten self-selected participants with no
screening question, so no market-size claim is made.

The screen-level drop-off names the step worth fixing, where the completion average hides it, and stating the limit stops the report from turning a locator into a verdict. The usual failure is treating a link-shared sample as representative and reading the headline rate as the result.