Measuring

Performance

How many extra conversions Optimeleon produced, per page and per variant, measured against the base arm, with the uncertainty shown rather than hidden.

Performance answers one question: how many extra conversions did Optimeleon produce? Not how many conversions happened, but how many happened because of the variants, measured against your original.

Impact

The headline is the extra conversions attributed to Optimeleon over the window you have selected, with a confidence interval beside it and a curve showing how it accumulated day by day.

The headline, the interval, the curve and each page's own board all report the same quantity, so the numbers on this screen agree with each other by construction.

The curve re-shapes when you move the window

The points are contributions to the selected window's total, not independent daily estimates. The figures are computed over the whole window at once, so changing the window legitimately changes the shape of the curve as well as its endpoint.

The confidence interval is not decoration

An interval that comfortably clears zero means the effect is real. An interval that straddles zero means you do not yet know, regardless of how good the headline looks. Optimeleon shows it rather than rounding it away.

The Board: per-page, per-variant

Drill into a page and you get its roster: each arm, including your original, with its adjusted conversion rate, its share of traffic, and a role.

Role	Means
Your original	The base arm. Everything is measured against it.
Portfolio winner	The single best arm on this page.
Winning	Confidently ahead of the base.
Ahead	Ahead, but not yet at the confidence bar a “Winning” verdict needs.
Specialist	Not the best overall, but strong for a particular slice of traffic.
No clear effect	Neither better nor worse than the base, as far as the data can tell.
Losing	Confidently behind the base.
Unclear	The data is confounded: something about this arm's exposure makes the comparison unreliable.
Still learning	Not enough conversions yet for a trustworthy verdict.

“Still learning” means wait

An arm needs a few dozen conversions of its own before its verdict can be trusted. Below that the numbers move around a lot and mean very little. Resist reading a winner into a variant with nine conversions. This is where most optimization programs go wrong.

A Losing verdict is not a permanent row on this table. Arms that have clearly been beaten are retired rather than left running, which frees capacity for the next variant you publish. Seeing one disappear from the live line-up is the system working, not data going missing.

Filtering by event

If your project has several conversion events, the event filter switches which one the whole screen is measuring. It defaults to your dominant event, so the headline is never quietly reporting a minor one.

What it does not tell you

Unconsented visitors. Measurement is consent-gated, so the absolute counts here are lower than your total traffic. The comparison is unaffected, because consent does not depend on which variant someone saw, so the effect size is unbiased even though the sample is smaller. See Consent & cookies.
Revenue you did not send. If your conversion event carries no value, Optimeleon can count conversions but not money.

Tip

If a number here disagrees with your own analytics, the usual causes are a different attribution window, a different definition of the conversion, or the consent gate. Bring the two definitions side by side before assuming either tool is wrong.