ThredUp uses AI to sort 80K daily items while cutting teams to just 4 (humans)

ThredUp, one of the world’s largest online apparel resale platforms, is using artificial intelligence to manage the 70,000 to 80,000 items that flow through its platform daily. The company has moved beyond basic analytics to deploy generative AI across both customer-facing features and backend operations, demonstrating how even digital-native companies must evolve their AI strategies as they scale.

The big picture: ThredUp processes over 100 million unique SKUs and has been using machine learning in production since 2015, but generative AI has transformed how the company handles both search functionality and operational sorting.

How it works: The company overhauled its search engine 18 months ago to enable AI-powered visual search that interprets product images rather than relying solely on database tags.

• Previously, searching for “Madewell Jeans” would return an overwhelming 50,000 results with basic taxonomy-driven filters.
• Now customers can search for “ugly Christmas sweater” and get relevant results even though those exact terms don’t exist in the database.
• The AI models help interpret visual elements and context to deliver more targeted results.

Operational efficiency: Generative AI models excel at category detection and style identification, helping sort through countless clothing brands, sizes, and categories automatically.

• The technology has reduced project team sizes from large cross-functional groups to teams of just four people.
• Previously, teams required data scientists, data engineers, front-end engineers, mobile engineers, and other specialists working in coordination.

Talent strategy shift: ThredUp is seeking employees with different skill sets as AI changes how work gets done.

• The company prioritizes candidates with a “growth mindset” toward AI rather than highly specialized technical skills.
• “You don’t have to be an expert at everything, but we want you to be curious, capable, and versatile,” said Dan DeMeyere, chief product and technology officer.

What they’re saying: DeMeyere emphasized how scale necessitates AI adoption: “When you get to a certain size, certain things don’t just scale manually. Rule-based systems, very simple algorithms just can only go so far.”

• “We want our product managers to be prototyping with AI—not doing all the prototypes by hand. And we want the engineers to take a prototype and get AI’s help and scaffold it out really quickly.”

Timeline: ThredUp has been running generative AI in production for about 20 months, building on nearly a decade of machine learning experience since launching in 2009.