Wayfair Uses AI to Catch and Correct Bad Product Listings

Source Domain: www.pymnts.com

Its catalog spans about 30 million items across nearly 1,000 product types. Roughly 47,000 attributes cover details such as color, material and dimensions. Those details determine how products show up in search results and recommendations. Before employing artificial intelligence (AI), fixing bad listings depended mostly on suppliers and shoppers flagging errors. Manual review couldn’t keep up with a catalog that size.

Through a partnership with OpenAI, Wayfair has corrected 2.5 million product attribute tags and now automatically handles 41,000 supplier support requests a month, the AI startup shared in a March post. The work covers two jobs that used to depend on people: merchandising staff checking product listings for accuracy and supplier support teams sorting through requests one at a time.

Wayfair built a system that reviews product listings against a description of what each detail is supposed to mean, then flags or corrects mistakes. If a coffee table is listed as walnut but the photo and description point to pine, the system catches it.

The system is running on more than 1 million products so far, OpenAI added. A test on corrected listings showed they got more clicks and ranked higher in search than before, according to OpenAI.

Wayfair didn’t let the system make changes unchecked. Staff manually inspect samples of the corrected listings. When the system is confident in a fix, it updates the listing and tells the supplier. When it isn’t, it asks the supplier to confirm the change first. Wayfair said it expects to expand the program to cover four times as many products within six months.

Advertisement: Scroll to Continue

PYMNTS reported that Wayfair’s use of Google’s AI models to categorize products had already cut the time needed to curate listings by 67% and lifted some conversion rates by 2%.