Choose the objective
Select balanced relevance, product discovery, average order value, or inventory velocity. The objective affects ranking priorities, while exclusions, inventory thresholds, and pins remain in force.
Choose the control level
Use Manual to make every change, Recommend changes to review suggested adjustments, or—on Growth and Scale—Automatic to promote supported experiment winners after the evidence threshold is met.
Run a ranking experiment
Choose distinct control and challenger strategies and allocate traffic. Storefront sessions remain consistently assigned while the app measures impressions, orders, and revenue per view.
Wait for meaningful evidence
Recommend does not promote a challenger from a small sample. Each arm requires meaningful traffic and a material revenue-per-view advantage; otherwise the experiment continues or retains the control. Promoted decisions remain in history and can restore the previous strategy.