Skip to content

Experiments

Test the page, decide by the sale — not by the click

The experiment engine runs on the pixel already installed. And because the platform knows who the person is, a purchase that lands days later goes back to the right arm of the test.

No extra script

Same pixel. No new tag, no additional tool to maintain.

Winners by revenue

The goal can be an approved checkout purchase matched by identity — something almost no testing tool can do.

Honest statistics

Minimum sample, minimum runtime and a traffic-split check before declaring a winner. No winners off ten visits.

How it works

A person enters one arm and stays there

Assignment is stable per visitor: whoever landed in variant B comes back to variant B tomorrow. Variants change the page through DOM operations or a redirect, and exposure is only counted once the change actually happened on screen — "exposed but not applied" is the classic defect that biases tests toward no effect.

  • Stable assignment, independent across experiments
  • Targeting by page, device and URL condition
  • Preview by link, without launching to everyone
  • Exposure counted only when the variant was applied
Experiment · sales page
A · control1,28%

2.418 exposed · 31 attributed sales

B · variant1,83%

2.402 exposed · 44 attributed sales

96% chance that B beats A. 3 more days of sample before the verdict.

Measurement

Tuesday's sale counts for Friday's test

The goal can be an on-site event — a lead, a checkout start — or a purchase recorded in your payment gateway. In the second case attribution uses the same identity graph: a purchase made eight days later on another device goes back to the arm the person was exposed in.

Decision

The verdict waits; the numbers do not

Until there is enough sample and runtime, the platform shows every number and withholds the verdict, telling you why. That is the opposite of the usual pattern: declare a winner early, reverse it next week, and lose the team’s trust in testing altogether.

  • Probability that the variant beats the control
  • Traffic-split check before any conclusion
  • Correction when there are more than two variants
  • An explicit reason whenever the winner is withheld

Frequently asked questions

Do I need a developer to create a test?

Not for swapping text, hiding an element or redirecting. For structural changes the recommended pattern is that the engine marks the variant and your site reacts to it — the developer sets that up once and never again.

Will the page flicker?

The decision is cached locally, so from the second visit on it is synchronous. On the first, the affected area is hidden briefly with a timeout that always releases.

How many visitors do I need?

It depends on the effect you want to detect, and the platform tells you how much is missing. As a rule, a test that needs purchases takes longer than a test that needs clicks.

Start measuring what actually becomes revenue

Create a free account, connect your first source and see your first lead timeline today. No card required.