Wayfair and OpenAI: Revolutionizing Customer Support and Catalogs

Wayfair and OpenAI: A Strategic Alliance for Operational Optimization

Wayfair, a major player in the home goods retail sector, has integrated OpenAI's models into its internal systems to transform its critical processes. This partnership, initiated through small-scale testing in 2024, has evolved into a comprehensive production system. The goal is to reduce manual labor, accelerate decision-making, and improve data quality across millions of products.

Rather than a one-off approach, Wayfair has opted to embed OpenAI's generative AI into its core operational workflows. The company has focused on areas where complexity and scale are crucial, including routing and resolving supplier support requests, as well as continuously improving product attributes in a catalog that contains approximately 30 million items.

Enhancing Catalog Quality at Scale

Wayfair's catalog management team is responsible for tens of millions of products spread across nearly 1,000 different categories. Accurate attribute labels, such as color, material, or size, are essential for search, recommendations, and merchandising.

Jessica D'Arcy, Associate Director of Catalog Merchandising at Wayfair, emphasizes that data quality is crucial for building trust with customers. Good data quality enables buyers to make informed purchasing decisions, thereby reducing costly issues like returns of poorly represented products.

Before the integration of OpenAI, improving labels relied primarily on reports from suppliers and customers. However, the volume was such that manual effort could not keep up. While early custom AI models for specific labels were effective, they were costly to develop and maintain. Carolyn Phillips, a machine learning scientist at Wayfair, explains that although technically functional, these models could not scale to cover the 47,000 labels needed.

Developing a Reusable AI Architecture

To overcome the limitations of single-use models, Wayfair established a label-independent system based on a unique OpenAI model. A "definition agent" ingests definitions from the web and internal sources to provide context for each label.

Phillips notes that the real obstacle was not the model's performance, but the human time required to define and encode each label. This context, combined with aggregated data from Wayfair's ecosystem, feeds a framework capable of classifying attributes across product classes. The team has thus been able to expand model coverage to new attributes at a pace 70 times faster than a year ago. The system is now in production for over 1 million products.

Measurable Impact on Teams

Since integrating OpenAI's models, Wayfair has seen significant improvements. In the catalog area, the company has reduced the number of incorrect or missing labels, correcting 2.5 million labels for the most visible and purchased products. They plan to quadruple this impact in the next six months.

For supplier support, triage, co-pilot, and self-pilot systems have automated 41,000 tickets per month, achieving up to 70% automation in certain workflows. This has reduced processing times by eliminating routine manual work, thereby improving supplier satisfaction and decreasing ticket reopenings.

The increased visibility of models on tickets and supplier intent, beyond what a single associate can see, has contributed to this rise in satisfaction.

Wayfair has also deployed over 1,200 ChatGPT Enterprise seats among its approximately 12,000 employees to support ad hoc tasks, internal problem-solving, and experimentation with generative models.

A Sustainable Partnership with OpenAI

Wayfair has a long tradition of investing in machine learning and collaborating with AI platforms and LLM providers to advance its business. Advances in cutting-edge models, including multimodal systems, are expanding creative possibilities for its teams. This is particularly relevant in home retail, where products are often visual and subjective.

Carolyn Phillips expresses her enthusiasm for the new challenges the company can now tackle. Traditional algorithms require strictly defined datasets, whereas these new models allow for working through ambiguity and context in a scalable manner.

Demand for ChatGPT Enterprise is strong among Wayfair employees, who see it as a practical tool for moving faster. Customer expectations are also evolving, with more buyers becoming familiar with AI in their daily lives and expecting similar capabilities when shopping online.

Fiona Tan, an executive at Wayfair, highlights that often, customers do not have the exact words to describe what they are looking for. Natural language and multimodal systems help bridge this gap.

For Wayfair's leaders, the goal is to enhance human expertise while developing internal capability. "We are building for a world where AI is an integral part of the shopping journey — whether on our site, through support, or via conversational interfaces," concludes Fiona Tan.