Platform · Learnings

Every test makes the next one smarter

Learnings reads every finished test on your website, finds what they show together and suggests what to test next. Code computes every number. The AI only puts the findings into words.

Knowledge that compounds

What your tests already know

A card for every finished test, patterns pooled across them, a map of what has never been tried, and a chat that answers from your own evidence.

  • A learning card for every finished test: what changed, what it taught and whether the hypothesis held
  • Patterns across tests, each with an honest evidence level
  • Statistics adjusted for the many patterns checked at once, so a lucky streak is not presented as a rule
  • Opportunities that open as a pre-filled draft experiment in one click
  • Answers drawn from your own tests, never from generic best practice
Per websiteOpt-inAI credits
The Learnings page for one website: a summary of what the tests show right now, then the strongest patterns, each with its evidence level, its estimated lift drawn as a range around zero, and the number of tests behind it.

Learning cards

One card per test, written when it ends

When a test concludes, Pertento reads the changes each variant actually made and writes a short card about it. The team’s own debrief sits beside it, always labelled as the team’s view, so a reading of the result is never mistaken for the result.

  • Each variant described in plain words, read from the changes made in the test
  • The outcome, and what the test taught in one sentence
  • Whether the hypothesis held: supported, contradicted, or the test says nothing either way
  • A possible reason and a suggested next test
A learning card for one finished test on a sign-up form: each variant described in plain words, a win of 4.8%, the lesson in one sentence, the hypothesis marked as supported, a possible reason, a suggested next test, its labels, and the team’s own debrief marked as the team’s view.

Patterns

What several tests show together

A single test proves very little. Pertento pools the tests that share something, such as the same kind of change on the same page, reports what they show together, and says how far that is worth trusting.

  • Five evidence levels: consistent across three or more tests, early signal, small effect, not enough evidence yet, or mixed
  • When two patterns are really the same tests seen twice, it says so instead of counting extra proof
  • Tests with a known problem, such as a tracking issue, are left out unless you include them
  • Filter by device to see what works on mobile on its own
The page for one pattern, reassurance beside the main call to action, rated consistent: a pooled lift of 3.1% across four tests with a range that stays above zero, and each contributing test listed with its own result.

Evidence

What has been tested where, and what never has

The Evidence view lays every page type against every type of change. Each cell carries the pooled lift and the number of tests behind it, and an empty cell is ground nobody has tested yet.

  • Page types against change types: layout, copy, navigation, trust, forms and more
  • A cell resting on one test is marked as one, so a lone result never reads as a trend
  • Every combination nobody has tried, listed as a gap
  • Each test labelled by change type, page, element and expected mechanism
The Evidence grid: page types across the top and change types down the side, each cell carrying a pooled lift and its number of tests in green or red, cells resting on one test drawn with a dashed edge, and blank cells marked untested.

Opportunities

From result to next test in one click

The evidence raises questions, and Opportunities turns each one into a concrete next test. Draft experiment, after a confirmation, creates a new draft that already carries the reasoning, so nobody has to rebuild it from memory.

  • Grouped by purpose: explain a surprise, replicate, resolve a mixed picture, isolate, or fill a gap
  • The draft opens with the hypothesis and the expected mechanism already written
  • Page targeting taken from the tests behind it
  • Links to the evidence, and the current estimate as the baseline to beat
An opportunity marked replicate: the open question, why it is still open, a suggested experiment and its expected mechanism, with Dismiss and Draft experiment buttons, beside a list of what the new draft carries over.

Ask about your learnings

Put a question to your own evidence

A chat built into the Learnings page answers in plain language from this website’s tests and nothing else. Every claim names the tests it rests on, and each reference opens that test’s details.

  • Ask “What hurts sign-up?” or “What should we test next?”
  • It looks up a test’s measured results when a question needs real figures
  • The conversation is shared with the team and saved
  • Ask about this experiment, on every experiment page, now knows what earlier tests showed as well
The Ask about your learnings panel: the question “What hurts sign-up?”, and an answer in plain words that names the three tests it rests on, warns that one finding comes from a single test, and suggests what to test next.

Foundation check

Learnings are only as good as the tests behind them

The foundation check shows how well past tests were documented: a clear hypothesis, an expected mechanism, a debrief. Better documentation means sharper learnings, and the check nudges the team towards better testing habits.

  • Hypothesis, description, mechanism and debrief, checked for every test
  • A gap marked missing, or weak when boxes were ticked and nothing was written
  • Fix opens the test where the gap is
  • Writing learnings for past tests is a separate step that shows its cost first
The foundation check: a table of past tests with a column each for hypothesis, description, mechanism and debrief, every cell marked ok, weak or missing, and a Fix button on each row.

Built not to overclaim

Learnings you can trust

This is a product that says “not enough evidence yet” rather than oversell a result. These are the rules that keep it honest.

Code computes every number

Lifts, ranges and evidence levels are computed by code from the measured results. The AI reads those figures and puts them into words. It never produces a number of its own.

Cautious by design

The statistics are deliberately sceptical, so a pattern has to earn its evidence level rather than be handed one by a single good result.

Adjusted for many checks at once

Checking dozens of patterns at the same time makes a few of them look strong by luck alone. The ranges are adjusted for that, so a lucky streak is not presented as a rule.

The same tests are not counted twice

When two patterns rest on the same tests, Pertento says so, instead of presenting one result seen from two angles as two pieces of proof.

Problem tests left out

A test with a known problem, such as a tracking issue or a site release during the run, is left out of the pooling. The team can choose to include it anyway.

Honest about a small programme

It works with a few tests and becomes more useful with every one that finishes. With only a few, it says how little they can prove rather than stretching them into a strategy.

Running it

How it fits your programme

Learnings is switched on per website, reads only that website’s tests, and is paid for in the account’s AI credits.

Off until you switch it on

Learnings is opt-in per website and off by default. Once it is on, it covers every test that finishes from then on.

Past tests, when you choose

Writing learnings for tests that finished earlier is a separate step you confirm, and it shows what it will cost before it runs.

Your own evidence only

Everything comes from that website’s own tests, never from generic best practice. An agency’s clients each keep their own learnings, separate from one another.

One record for the whole team

Cards, patterns and the conversation in Ask about your learnings are shared and saved, so what the programme has learned survives people moving on.

Tell it what you already knew

Mark a pattern as already known or not useful, and the page adapts to what your team has told it.

Paid for in AI credits

Writing cards, Ask about your learnings and the experiment chat use the account’s AI credits. Accounts not billed by card have a monthly AI allowance instead.

Questions

Before you switch it on

How is this different from asking a chatbot about our results?

In two ways. The numbers are not the AI’s: lifts, ranges and evidence levels are computed by code and adjusted for the number of patterns checked at once, and the AI only describes them. And the answers come only from this website’s own tests, with every claim naming the tests it rests on, so you can open each one and check. A general-purpose assistant can sound confident about anything. This one is built to say “not enough evidence yet” when that is the honest answer.

Will it tell us which variant is going to win?

No, and it is built not to. It summarises the evidence your finished tests produced and suggests what to test next. It does not predict winners or guarantee a result, and when the evidence is weak it says so in as many words.

We have only run a handful of tests. Is it worth switching on?

It works with a few, and it is honest about how little a few tests can prove: most patterns will read as an early signal or as not enough evidence yet, which is the correct reading. The value grows with every test that finishes. The foundation check and Opportunities are useful from the start, because they show what to document better and what to test next.

Does it learn from other companies’ tests?

No. Learnings is strictly per website: it reads that website’s tests and nothing else, and it does not compare you with other sites or with an industry benchmark. For an agency, each client’s learnings are kept separate.

What does it cost?

It uses the account’s AI credits: writing learning cards, Ask about your learnings and the experiment chat all draw on them. It is off by default, and once switched on it covers tests that finish from then on. Writing learnings for past tests is a separate step that shows its cost before you confirm it. Accounts not billed by card have a monthly AI allowance instead.

What language are the learnings written in?

English. Learning cards, patterns and opportunities are written in English, whichever language your team reads this page in.

Everything in the box

The full feature set

Experimentation

Statistics

Monitoring

Voice of customer

Learnings

  • Learning cards
  • Patterns across tests
  • Five evidence levels
  • Adjusted for multiple comparisons
  • Same-test patterns merged
  • Problem tests left out
  • Results by device
  • Evidence grid
  • Opportunities
  • Draft experiment
  • Ask about your learnings
  • Foundation check
  • Mark as known or not useful

Pertento is a conversion-rate-optimisation platform for websites and online stores, built for both in-house teams and CRO agencies running experimentation programmes across a client roster.

Integrations

Works with the stack you already run

One snippet drops into any site, or go fully server-side. Nothing to re-architect.

One-line snippet

Paste one tag in your head and you are live: a 0.9 KB loader, no build step, no dependencies.

Google Tag Manager

Deploy and manage experiments straight through GTM. No developer time required.

Server-side API

Run tests beyond the browser: pricing, search and routing, flicker-free.

  • Shopify
  • BigCommerce
  • Shopware
  • Centra
  • Geins
  • Saleor
  • WordPress
  • Optimizely CMS
  • Storyblok
  • Next.js
  • Vue
  • Astro
  • Matomo
  • Amplitude
  • RudderStack
  • Slack
  • WooCommerce
  • Salesforce
  • Wix
  • Litium
  • commercetools
  • Medusa
  • Drupal
  • Contentful
  • Webflow
  • Nuxt
  • Angular
  • Remix
  • Piwik PRO
  • Mixpanel
  • Tag Manager
  • Webhooks
  • Adobe Commerce
  • PrestaShop
  • Squarespace
  • Norce
  • Shopify Hydrogen
  • Vendure
  • Umbraco
  • Sanity
  • Framer
  • React
  • SvelteKit
  • GA4
  • Adobe Analytics
  • Segment
  • Klaviyo

Not on the list? If it renders HTML, Pertento can test it:read how each one connects.

Better experiments, better conversions

Test on all visitors with the world’s lightest script and make confident decisions powered by real-time reporting.

Start trialRequest a demo