Reading the Learnings Page

Once learnings are on and the first ones are written, Experiments → Learnings opens on three views. Everything on the page belongs to the website you have selected.

The row above every view

• The views, as tabs on the left: Learnings, Evidence and Opportunities.
• N/N experiments, on the right: how many of the website's finished tests have a learning.
• Ask about your learnings, a field that opens a chat beside the page. See Ask About Your Learnings.

When you open a pattern or a test, the tab you came from stays selected and a trail under the row links back, for example "All learnings › Trust on checkout › #5871". The browser's Back button steps back through the views too, from a test to its pattern and from a pattern to the list, before it leaves the page.

Learnings

The view you land on.

• The headline: what your tests add up to, in a sentence or two.
• Key learnings: the strongest patterns, one for each distinct set of tests, each with its estimated lift and a range.
• The evidence map: every pattern on one axis, with its range drawn as a band, so you can see at a glance which patterns clear zero and which do not.

The page also says how many patterns are conclusive so far. How Learnings Weighs the Evidence explains the ranges and the evidence levels.

Evidence

Every pattern, grouped four ways:

• by page and change type, for example layout on a pricing page;
• by mechanism, the reason the test expected the change to work;
• by theme, which gathers tests chasing the same idea under different names;
• by element, the part of the page that changed.

The view also lists combinations of page and change type that have never been tested, and what is not counted yet: tests left out, tests with no expected mechanism, and tests with no usable range. The foundation check is linked from here.

Opportunities

The open questions your tests raise, each with a Draft experiment button. See Opportunities in Learnings.

A pattern

Open any pattern to see:

• its estimate test by test, so you can see which tests carry it;
• what the tests show, in words;
• a possible reason, written by AI and marked as such: it is a suggestion, not a measurement;
• a next step.

Two buttons tune what the page highlights:

• Mark as known, for a pattern your team already knew.
• Not useful, for one that does not help you.

Your choice is shared with everyone working on the website, and you can undo it. The headline and highlights are reworded at the next hourly update. Neither button changes a number: the pattern's estimate, range and evidence level stay exactly as the statistics computed them.

A test's debrief

Open a test from a pattern, from Ask, or with Open debrief in the Library. You see:

• what each variant changed, and what the test taught;
• the team's read from the debrief beside the AI's learning, and whether the two agree;
• what the test has written down;
• which patterns it counts in.

Until someone checks it, each learning says "Written by AI from the test's changes and result. Nobody has reviewed it." Two buttons are here for roles that can create experiments:

• Mark reviewed, once someone has checked the learning. If the learning is rewritten later, it needs reviewing again.
• Tracking is fixed, include it, for a test left out because its debrief reported a tracking issue or a site release. See How Learnings Weighs the Evidence.

The device filter

The device filter pools the evidence again for one device: the tests aimed at it, and the tests that ran on every device. Each pattern's estimate, range and evidence level are recomputed for that device. The headline, the key learnings and the opportunities always cover every device.

Learnings elsewhere in Pertento

• The Library. Open a finished test's row to see its learning: What changed, What it taught, The team's read, and notes on the evidence, with an Open debrief link. You can filter the Library by change type, theme and element.
• Drafts. While you write a draft's hypothesis, similar past tests on the website appear under it, for example "#412 tested this in March: lost, -2.8%" or "#455 is running on this page now". They are found by matching pages and words, not by AI, and use no credits. A test that only counts as a win because its variant was deployed shows as "won (deployed)".
• Ask about this experiment. The assistant on an experiment's page reads the website's learnings when it answers.