Retail / E-commerce
Wayfair: AI validation and tagging of product attributes in the catalog
- Company
- Wayfair
- Country
- United States
- Adoption stage
- In operation
- Source published
- Date basis
- The date the source was published. It can differ from the date adoption started.
- How the source was checked
- Read the full source text
The work problem
According to a Wayfair customer story published by OpenAI on September 25, 2026, Wayfair manages a catalog of roughly 40 million items across nearly a thousand product classes, and tags such as color, material and size power search, recommendations and merchandising. Tag errors used to be found only when suppliers or customers reported them, manual effort could not keep up with the volume, and early custom AI tagging models proved expensive at scale. The company had 47,000 tags to support.
Technology and data
Wayfair built a catalog quality system with OpenAI models. The system first uses web information and Wayfair's internal definitions to create a guide explaining what each tag means, and the models then use that guide to review product information and assign the appropriate tags. With an audit process in which associates inspect product samples and with supplier validation, high-confidence changes update product content directly and notify the supplier, while low-confidence or high-risk changes take effect only after supplier confirmation.
Results
According to the story, the AI systems Wayfair built with OpenAI models validated more than 11 million product specs, and the catalog quality system has processed more than 1 million products. Wayfair is expanding the range of product details its AI can check at 70 times last year's pace, and a controlled A/B test showed a significant increase in impressions, clicks and page rank in the treatment group.
Limits and open questions
The story does not disclose the duration of the A/B test, the size of the increases or tag accuracy figures.
Sources
- Wayfair boosts catalog accuracy and support speed with OpenAIopenai.com, Accessed
Compiled from public sources. These are not results from ATF Works customers.