A campaign can hold more than one version of the popup. Each version is a
variant, and once two or more of them are switched on, your traffic is split
between them — that is the A/B test.
Where variants live
Open a campaign and stay on the Overview tab. Under the metrics you will find
the variant table, with one row per variant.
The desktop table has these columns: Traffic, Variants, Active, three
performance columns that follow the campaign's goal (for the default goal:
Impressions, Submits, Submit rate), and Uplift.
Each row shows a thumbnail of that variant, its name, its traffic share as a
percentage, and a switch that turns it on or off. While a variant is still being
built, the thumbnail is a working placeholder and the row says
Building · step {step} of {total} instead of showing a preview.
Adding a variant
Under the table there is Add A/B test variant. It opens a menu with three
options:
Control (no popup) — "A holdout bucket that sees nothing". This is the
slice of visitors who get no popup at all, so you can measure the popup against
showing nothing. You can only have one, and the menu says Already added if
you do.Duplicate a variant — "Clone an existing design to tweak". Pick one of
your existing designs; you get a copy named after it, with its own build, ready
to edit.Create with AI — "Generate a new design in the designer". Disabled when
the campaign has no design session: "Unavailable — this campaign has no design
session".
Two more routes exist. Each row's own menu has Rename, Duplicate,
New design from this variant and Delete. And from a design session's
concept board you can select finished concepts and use Add {count} as A/B
variants to attach them to an existing campaign.
Whenever you add or remove a variant, traffic is redistributed evenly across
whatever is active.
Asking the assistant for test ideas
The chat next to the campaign can propose variants. Ask something like "Suggest
an A/B test I could run" — it is one of the suggested prompts. The assistant may
first ask a couple of questions, then come back with a set of labelled ideas,
each with a preview.
Click an idea to open Variant comparison: your Control · current on one
side, the Suggested version on the other, plus Why this works and
What changes. Add variant creates it in the campaign; the button then
reads Added. If none of them appeal, there is None of these — show me other
variants.
Editing a variant
Click a variant row and it opens in an overlay: desktop and mobile side by side,
with Preview and Edit with AI tabs. Describe a change and it rebuilds
that variant.
Nothing here reaches your visitors on its own. Every change is kept as a draft —
the header shows an Unpublished draft badge — and "Nothing changes for your
visitors until you press Save." Save is the control that publishes.
For hands-on work there is Drag & drop editor in the overlay footer. If two
variants still share one build, opening the editor forks a copy first, so editing
one variant never changes the other.
Running the test
There is no separate "start test" button. Once two or more variants are active on
a live campaign, a banner appears above the table saying A/B test running.
Until there is enough data it reads Data collection in progress.; after that
it names the leader — "{name} is leading by +x.xx% (nn% chance to win)".
The banner also carries Declare winner at {threshold}%. Pressing it keeps the
leading variant serving and switches the others off. If the leader's chance to
win is below your threshold, or there is no data yet, you get a confirmation
first — Declare winner now? with Keep testing and Declare anyway.
Clicking the Traffic cell opens A/B test settings, which holds
Auto-declare winner with Declare at … chance to win, Min conversions
and Min days running, plus a Traffic share section. Traffic share is
Evenly split, marked Always on: "Traffic is shared equally across all
active variants." A custom split is not available yet — the panel says
"Custom traffic split is still in progress."
Where the results show up
The variant table itself is the scoreboard: impressions, conversions, conversion
rate and Uplift per variant, with the chance-to-win percentage next to the
leader. Above it, the metrics band shows the campaign totals for the goal and
date range you pick. The campaign also has an Analytics tab.
One thing to know about the numbers: the per-variant figures are all-time totals,
and uplift is measured against a baseline variant — the holdout if you have one,
otherwise the oldest variant. Chance to win is a Bayesian estimate, so it needs a
reasonable amount of data before it means much.