A/B Testing

Current availability

The current Analytics > Experiments page (/app/analytics/experiments) displays existing experiment records, or an empty state when none are returned. Merchant controls for creating, configuring, stopping or applying an experiment are not available in this interface.

Earlier versions of this guide described a creation modal, a per-experiment results view and an apply-winner action. Those instructions did not match the current page. This correction does not establish that all experimentation infrastructure is absent: backend list, create and end endpoints exist, but they do not establish the complete shopper-assignment workflow described by the old guide. They are not a supported public API.

Do not expect an experiment-creation control to appear simply because your store reaches a traffic threshold. Confirm an accessible, supported workflow with the team before planning an experiment.

Reading the current page

1. Open FilterIQ in Shopify admin and navigate to Analytics > Experiments.
2. If no records are returned, the page shows an empty state. This does not create or schedule a test.
3. Where records exist, the table shows Name, Status, Metric and Traffic Split.
4. Treat these as record fields. A running status alone does not prove that shoppers receive different configurations or that event collection is complete.
5. If a record is unexpected, share its name and the observed state with support. Do not send access tokens or shopper information.

The page does not currently expose per-variant results, confidence intervals or a control to publish a winner. Do not infer an improvement in conversion from its traffic percentage or status badge.

Prepare a useful evaluation

You can still define an evaluation brief before a supported experiment is agreed:

• State one shopper problem and one proposed change, such as the ordering of two existing filters.
• Preserve the current configuration and a tested rollback procedure.
• Define the population, assignment method, event definitions and reporting period before collecting evidence.
• Check tracking consent and data coverage. Missing events are not zero activity.
• Agree the sample-size and analysis method before starting; a short run or a higher observed rate alone does not establish a reliable improvement.

For ordinary diagnosis, use the analytics guides and reproduce the query with its active filters. A before/after observation can guide an investigation, but does not isolate the effect of a configuration change.

Next step

Contact the team to confirm which evaluation workflow is available in your private preview. Building a complete merchant-managed experimentation experience is separate product work; this guide does not promise its availability.