Once an A/B test is running, Instant shows results tied to your primary goal, per-variant performance, and a confidence score. This guide helps you read those numbers and decide what to ship.
Open test results
In the project sidebar, select A/B testing.
In the list, each test shows its Goal, Started date, Confidence, and a Status badge (Active or Done).
Select a test to open its results.
Results update as new sessions arrive. Refresh after major traffic events such as an email blast or a paid-media spike.
The winner card
At the top of a test, Instant shows whether there is a winner yet:
No winner yet — the test is running but confidence isn't high enough.
No winner — the test ended without a confident result.
The winning variant's name — once confidence passes the threshold.
Alongside it you'll see Confidence (how sure Instant is about the result). While there's no winner, it also shows Total sessions; once there is a winner, it shows the winning variant's value for your selected goal.
You can change the Goal from the results page — Instant recalculates the winner for the goal you pick.
The results table
Each variant is a row you can sort. Columns:
Column | Meaning |
Variant | The variant name (the first variant is the Control when you have three or more) |
Sessions | Unique visitors enrolled in that variant, and its share of test traffic |
Conversions | Sessions that resulted in a completed purchase, and the conversion rate |
Revenue | Total sales revenue from that variant's sessions, based on completed orders |
RPV | Revenue per visitor — total revenue divided by total sessions |
CTR | Share of visitors who clicked through to a page other than the variant's URL |
AOV | Average order value — total revenue divided by the number of conversions |
Each row links to the variant's storefront URL, and the winning variant is marked with a medal.
How confidence works
Instant uses a Bayesian model to compare each challenger against the control for your selected goal. A variant is treated as a winner when its confidence is above 90% and the estimated difference is clearly on one side of zero.
Instant needs enough data before it names a winner: at least 50 conversions per variant (or 50 click-throughs when the goal is CTR). Until then, the test shows No winner yet, even if one variant is ahead.
Low-traffic destinations take longer to reach confidence. A product page with a handful of sessions a day may need weeks; a high-traffic page can resolve in days.
End a test and redirect traffic
There is no separate "declare winner" button — you end the test and choose where traffic goes:
On the test's page, click End test.
If confidence isn't high enough yet, Instant warns you can still end early. Choose End test to continue, or Keep running.
In the Ending test step, pick which variant to direct all traffic to (the winner is pre-selected when there is one), then confirm End A/B test.
After ending, all traffic through the A/B test link goes to the chosen variant. You can Change the redirect variant later, or Continue test to split traffic across all variants again.
Ending a test doesn't delete the other variants' destinations — clean those up in the builder if you no longer need them.
Inconclusive tests
No clear winner is still useful:
The change may simply not matter.
The effect may be too small to detect at your traffic level.
External noise (promotions, stockouts) may have swamped the signal.
Note inconclusive outcomes so the team doesn't retest the same tweak blindly.
Session pool effects
Enrollment draws from your monthly A/B test session pool — 10,000 on Pro and 100,000 on Business. If you run out, new visitors stop being enrolled until the pool resets next month or you upgrade (Upgrade your plan). Results for already-enrolled sessions stay valid.
After the winner ships
Keep an eye on Attributed revenue for the winning destination in Analytics overview.
Fold what you learned into your next hypothesis.
Use Instant AI and the Agent to build follow-up changes on the winning version (Edit with AI).
For setup, see Create an A/B test and A/B test goals and targeting.