Opportunities in Learnings

Learn what the five kinds of opportunity mean, how Draft experiment turns one into a draft with its hypothesis and targeting filled in, and how to dismiss one.

2 min read

The Opportunities view lists the open questions your finished tests raise. Code decides what counts as a question, from the patterns and the surprises, and the AI words it. There are five kinds, with at most three of each at a time.

The five kinds

  • Explain a surprise: a result nothing predicted. See "Surprises" in How Learnings Weighs the Evidence.
  • Replicate: a positive lean that the range cannot yet tell from zero. Another test would settle it.
  • Resolve mixed: tests of the same thing that disagree in direction. When the split lines up with device, the opportunity says so.
  • Isolate: one test that changed several things at once, so it is unclear which change mattered.
  • Fill a gap: a change type you have tried elsewhere, but never on a page you have tested at least twice.

The list keeps in step with your tests. New questions appear as tests finish, and a question the evidence no longer raises is retired.

Draft experiment

Draft experiment turns an opportunity into a new draft. After you confirm, Pertento:

  1. 1creates a draft with a hypothesis written from the question (using the figures the page showed), the expected mechanism, and two variants;
  2. 2copies the URL rules of the newest test behind the question, when those tests share a page;
  3. 3marks the opportunity as drafted, with a snapshot of the evidence as the baseline the new test has to beat;
  4. 4opens the new experiment.

Build the variants in the editor and check the targeting as you would for any other experiment.

The evidence is read again at the moment you draft. If the question has changed since you opened the page, the draft is refused, so reload the page and try again. Clicking twice still makes one draft.

Draft experiment is available to roles that can create experiments.

Dismiss

Dismiss hides an opportunity. It stays hidden while the evidence keeps raising the same question. Anyone who can see the website's learnings can dismiss one, Members included, and the headline is reworded at the next hourly update.