Size Chart or Personalized Size Advice A/B Test

Visitors who open the size chart and leave without choosing, sizing questions in support tickets, or fit problems that keep showing up in return reasons point to a hard size decision. If recommendation-first help leads, shoppers who fill in the form reach a size faster and form abandonment does not cancel out that gain; if chart-first help leads, the extra step costs more than it saves. If the chart itself has missing or wrong measurements, fix it first, and do not start the test until the recommendation method is validated for these categories and has enough data behind it.

Size selection completion

Compare size selections per eligible assigned visitor, including visitors who do not finish the recommendation form.

Also track

• Add-to-cart rateCheck whether the help format changes the first purchase step.
• Size-related return rateTrack fit-related returns for orders that complete the same follow-up period.
• Recommendation acceptance rateReport selection of the recommended size as a diagnostic measure in the recommendation arm.
• Input form abandonmentTrack visitors who leave the height and weight form unfinished.

• Help formatCompare a size-chart-first experience with a recommendation-first experience that still provides access to the chart.
• EvidenceKeep supporting claims and social proof unchanged.
• InputsKeep input fields fixed and optional.
• ModelValidate and hold the recommendation method constant for this comparison.
• CategoryInclude only categories supported by the recommendation method.

1. 1Compare the two on size selection completion per eligible assigned visitor, counting visitors who do not finish the recommendation form. Recommendation acceptance exists only in one arm, so treat it as a diagnostic. If selections rise while add-to-cart rate stays flat or the size-related return rate goes up, do not call the recommendation arm the winner.
2. 2Leave size ranges alone in this test; if the format and the ranges change together, you cannot tell which one caused the difference. The findings cover only categories the recommendation method supports, and returns should be compared for purchases past the follow-up period.
3. 3Do not present a recommendation built on thin data as if it were certain, since a wrong size skews the comparison as well. When the gap is unclear, use input form abandonment to see where the form loses visitors, and check whether the completion difference could be chance with the A/B test significance calculator.