How ABSee records experiments. And what “kept” means.

What counts as an experiment

An experiment is two versions of the same public page served to ordinary visitors at the same time: one visitor gets the original, another gets the version being tested. That split is the line ABSee draws. A page that simply changes, for everyone at once, is a change, not an experiment.

ABSee is experimentation intelligence for ecommerce: a record of real experiments run by 1,150+ brands, with the page before, the version tested, and whether it was kept. Today that record holds 10,000+ experiments, across home, category, product, cart, checkout and search pages.

What we record

ABSee reads what sites serve publicly to every visitor and records what its own ordinary visits receive. It looks only at public pages, visits at a respectful volume, and never uses accounts, logins or private data.

For each experiment the record keeps:

• the page before, as a visitor saw it;
• the version tested, as a visitor saw it;
• the brand, the page type and what changed between the two;
• when the experiment was first seen and when it ended;
• the outcome, once the brand has made its decision.

How an outcome is decided

The outcome comes from what the brand did next, read from the same public page:

• Running: both versions are still being served.
• Kept: the test ended and the tested version became the page.
• Rolled back: the test ended and the original stayed.
• No clear result: the test ended without a clear sign either way, for example when the page changed again for everyone.

The honest limit: this is the brand's decision, not a measured lift. ABSee does not see a brand's own numbers. A brand can keep a version for reasons beyond the result, such as brand guidelines or a wider redesign, and a rolled-back idea can still have been a good one for someone else. Outcomes are tracked and are part of the Intelligence plan.

What we leave out

• Redesigns and rollouts without a split. If everyone gets the new page at once, nothing was tested.
• Logged-in areas. Account pages, order history and anything behind a login are never visited.
• Prices personalised to accounts. Prices or offers that depend on who a customer is are not recorded as experiments.

How tool adoption is measured

The same public visits show which testing and personalisation tools a page loads in the browser. That is the data behind the experimentation platform market share report and the tool guides. The sample is brands in the ABSee record — mostly mid-to-large ecommerce, weighted to fashion and to Europe; only tools that load in the browser are visible, so server-side testing is undercounted.

Shares are published rounded, and grouped into bands from very common to rare. Raw counts are not published. The figures are refreshed quarterly from the same job, so one edition can be compared with the next.

Known limits

• Server-side tests. When a brand decides on its server which version to send, both versions can still be seen, but the tool behind them often cannot. Tests that change only what happens after a click may not show at all.
• Small brands. The record leans to mid-to-large brands. Smaller stores run fewer tests and are covered more thinly.
• Regional versions. A brand can test in one country and not another. ABSee sees the versions served where it visits, so a test limited to another region can be missed.

“Other people's tests don't apply to me”

That is right, and it is not what the record is for. Another brand's experiment is evidence for a hypothesis, not a result to copy. It tells you that a team with a similar page thought an idea was worth testing, what they tested, and whether they kept it. That makes your own hypothesis sharper and your test list shorter. Your audience still decides, in your own experiment.

Corrections and removal

If a record is wrong, or you run a site and want a page looked at, email hello@absee.it. We check every request, and we correct or remove what is wrong. How ABSee handles personal data is in the privacy policy.