How to Roll Out an A/B Test Winner at 100%
Serve a winning variant to 100% of targeted traffic while you make the change permanent — serving only, nothing recorded or billed, and you can stop any time.
Here is how to roll out an A/B test winner: serve the winning variant to 100% of the experiment’s targeted traffic, make the change permanent in your theme or site, then stop serving. When an experiment concludes with an honest winner — ≥95% probability to be best and no guardrail breach — you usually want visitors to keep seeing the winning variant while you hard-code the change into your theme. Winner rollout does exactly that: serve the winner at 100% of targeted traffic, from the pixel, until you’ve made the change permanent.
Serving, Not Measurement
A rollout is a serving mechanism, not an extended experiment:
- Every visitor on the experiment’s targeted pages sees the winning variant — no more bucketing.
- No events are recorded. The experiment is over; rollout traffic produces no new results.
- Nothing is billable. Rollout traffic never counts toward your plan’s sessions.
- Your theme is untouched. Stopping the rollout instantly returns visitors to your original theme.
- Theme and template winners keep being served the same way. Visitors on the targeted pages are sent to the winning theme or template through the pixel, at 100%, until you publish or assign it in Shopify.
Rolling Out a Winner
On a completed experiment with a winner, the Results tab shows a Keep serving the winner card under the verdict. Click Serve winner at 100%; a dialog titled with the winner’s name (“Serve … to everyone?”) names the pages it covers. Press Start serving, and the pixel starts serving the winning variant to everyone on the targeted pages. A pinned banner then shows Winner serving at 100% with the start date — and, if Autopilot did it, a Rolled out by autopilot chip linking to the autopilot log.
Only workspace owners can manage rollouts; other members see “Only workspace owners can manage rollouts.” The card appears only for a recorded winner with no guardrail breach.
Making It Permanent
A rollout is a bridge, not a destination. Make it permanent opens the Make this change permanent dialog. It summarizes exactly what the winning variant changes (CSS selectors, JS, or redirect — with a View code link; for a theme or template, Theme — Variant A — publish it in Shopify or Template — product.alternate — assign it in Shopify) and walks the three steps:
- Apply this change in your theme/site. For a theme winner: Publish it in Shopify before stopping, otherwise visitors return to your old theme. For a template winner: Assign it in Shopify before stopping, otherwise visitors return to the default template.
- Publish.
- Stop serving below — visitors then get your theme’s version directly (for a theme winner, your published theme).
A theme or template winner also shows a Make it permanent block on the Results page, under the rollout banner while it serves, with Publish in Shopify or Open products — see When to Stop an A/B Test and Declare a Winner. The serve dialog carries the same sentence before you start serving.
Press I’ve made it permanent — stop serving when you’re done, or Later to keep serving.
While a Rollout Is Serving
- Regression watch. When a north-star source is connected in Global Goals, Split Test Pro compares your north star’s daily average over the first 14 days of the rollout with the 28 days before it, from the same source. Once there are at least 7 days of rollout data, a drop of more than 10% sends one email and shows one banner: “Your site metric moved after this rollout — this is an observational signal, not proof the rollout caused it.” Check your north-star series, then decide whether to keep serving. The rollout is never reverted for you.
- Stop any time. Stop serving asks you to confirm, then visitors see your original theme again immediately. The experiment page keeps a record of the serving window (“Winner served at 100% from … to …”).
- Plan changes don’t break live rollouts. If your subscription changes, an active rollout keeps serving (serving is not gated) — new rollouts need Pro, and you’ll be nudged to make the live one permanent.
Autopilot Rollouts
With the Auto-rollout trust rung unlocked (after you’ve rolled out one winner manually), autopilot can serve a clean winner automatically when a test concludes. Autopilot rollouts follow every rule above, are logged, and are emailed to the workspace owner. See Autopilot.
Next Steps
- What counts as a winner in the first place: When to Stop an A/B Test and Declare a Winner
- The guardrails that can block a winner call: Global Goals
- The program that queues the next test: Testing Overview Coming soon
Frequently asked questions
What happens to an A/B test winner after the test ends?
Ending a test doesn't change your site: everyone sees the original page again. To keep visitors on the winner, serve it at 100% of targeted traffic with a rollout, then apply the change to your theme or site and stop serving.
How do you implement an A/B test winner permanently?
Copy the winning variant's CSS, JS or redirect into your theme or site, publish, then stop serving the rollout. For a Shopify theme or template winner, publish the theme or assign the template in Shopify before you stop serving, otherwise visitors return to the old one.
Does serving a winner at 100% count toward your plan?
No. A rollout is a serving mechanism, not an extended experiment: no events are recorded and rollout traffic never counts toward your plan's sessions. Stopping it returns visitors to your original theme immediately.
How long should a winner rollout run?
Only until the change is permanent: a rollout keeps serving until you stop it, but it is a bridge, not a destination. After 4 weeks of serving, the experiment shows a reminder banner and the workspace owner gets one email asking them to make the winner permanent. Making it permanent promptly keeps your site fast and independent of Split Test Pro.
Who can roll out an A/B test winner?
Only workspace owners can manage rollouts, and the rollout card appears only for a recorded winner with no guardrail breach. Once the Auto-rollout trust rung is unlocked, autopilot can also serve a clean winner automatically when a test concludes.