Skip to article
SectionMarketing

Marketing

A/B Testing

Updated

A/B testing stores suggested experiments, generated variations, and measured results. Review the target and the evidence before running a test or applying a winner.

Where to find it

In your dashboard, open the tool under Marketing. Open A/B testing.

Set it up

  1. Click Find opportunities to request suggestions based on restaurant signals.

  2. In Suggested tests, review the reason and hypothesis. Use Generate variations when needed, then inspect the control and alternatives.

  3. Choose Run test to start or Dismiss to leave the suggestion unused.

  4. In Settings, review Automatic testing and Applying winners. Click Save settings after making changes.

Daily use

Running tests show views, conversions, conversion rate, and sample progress. A conversion means different things for different test types: for example, a menu-description test counts adding to cart, not a completed paid order. Stop early ends a test before its usual duration; a small sample can leave the result inconclusive.

Completed tests can show Apply winner when an alternative won and has not been applied. The default winner-application mode is manual, even though automatic opportunity detection is enabled by default. Review your settings rather than assuming every result changes the site.

Application is target-specific. A winning menu description changes that menu item; a winning hero headline changes the restaurant tagline. An email-subject winner updates the newest draft or scheduled campaign, not an already-sent campaign. Some copy test types only save an override in restaurant settings; check your live store to make sure changes appear to customers as expected.

Common questions

Are winners applied automatically?

The default is manual review and application. You can change the winner-application mode in the test settings.

Does every test measure paid orders?

No. The conversion event depends on the test type. Menu-description tests count adding an item to the cart.

Can I stop a test early?

Yes, but a small sample can leave the result inconclusive. A projected impact is not a guarantee.

Limits

  • Projected revenue and confidence are estimates, not promised outcomes.
  • Nightly detection and periodic evaluation run automatically in the background.
  • Some experiment types may not appear on every customer-facing area of your online store.

Was this page helpful?

Help center

Still need help?

Call or email us and we’ll walk you through it.