With its ability to process immense datasets, deliver hyper-personalized recommendations and negotiate prices, generative AI is transforming retail into an ecosystem where innovation and convenience collide.
"AI isn't replacing jobs, but those who don't embrace it risk falling behind," ThredUp Chief Product and Technology Officer Dan DeMeyere told PYMNTS during a discussion for the series "When Chatbots Go Shopping: How GenAI Is Shaking Up the Retail Status Quo."
Unlike traditional AI, which can rely heavily on pre-programmed responses, generative AI creates interactions based on a combination of customer behavior, preferences and contextual data.
AI innovations are increasingly more than bells and whistles; they create sticky customer relationships that convert casual shoppers into loyal brand advocates. For retailers, the long-tail benefit lies in the wealth of data they can harness.
DeMeyere said he envisions a future where his own company's shoppers can craft entirely personalized experiences.
"Imagine composing your own thrift shop from scratch in seconds," he said, adding that by combining voice input, brand preferences and sizes, ThredUp ultimately aims to create one-of-a-kind, immersive shopping journeys tailored to individual users.
ThredUp, an online resale platform championing secondhand fashion, is itself tapping AI to redefine consumer experiences and operational models in secondhand shopping across five key areas: enhanced search; product discovery; dynamic pricing; hyper-personalization; and internal operations.
DeMeyere said AI can redefine customer engagement.
"We've trained a specialized generative AI model, a clip model, on fashion data, enabling customers to search for visual terms instead of text," he said.
By interpreting visual style language, the platform provides relevant results even for abstract terms. For instance, searching for "ugly Christmas sweater" yields precise results, although neither "ugly" nor "Christmas" appear in the database. This nuanced capability empowers shoppers to articulate their unique tastes effortlessly, DeMeyere said.
At the same time, discovery tools, like ThredUp's Image Search and Style Chat features, are crucial for engaging customers who may not have a specific item in mind. The company's Image Search lets users upload photos from sources like Instagram, identifying items within the image and matching them to its inventory. Style Chat enables shoppers to describe or imagine styles, and AI retrieves or "creates" matching items, he said.
Behind the scenes, the retailer uses AI to optimize pricing for its inventory and ensure a balanced flow of inventory and competitive pricing.
With over 4 million items listed and tens of thousands added daily, managing pricing manually would be infeasible, so the company gets help from its reinforcement learning model, DeMeyere said.
"When certain items aren't selling, the AI model automatically lowers prices, tracking real-time shifts in demand and revenue potential," he said.
ThredUp's own business model -- dealing with one-of-a-kind inventory -- adds complexity to AI implementation.
"Unlike Netflix or traditional retailers, our items disappear once sold," DeMeyere said.
This ephemeral nature complicates recommendations and customer decisions. To address these challenges, ThredUp developed "AI atomic functions," modular tools like garment detection and item description generation that streamline problem-solving, he said.
To enhance customer satisfaction, ThredUp also uses vector embeddings to understand individual shopper preferences.
"Whether customers are using core search or Style Chat, we tailor results to their unique tastes," DeMeyere said.
Personalized recommendations ensure the most relevant products rise to the top, increasing engagement and conversion rates.
From an internal standpoint, ThredUp's engineers use AI tools like Cursor AI and GitHub Copilot to enhance productivity, he said.
Looking ahead, DeMeyere predicted that down-the-line advances in AI, such as AI agents, could serve as behind-the-scenes stylists, ensuring customers never miss new arrivals.
"Agents could know a shopper's size, preferences and style, presenting the best items listed in the last 24 hours right on their app," he said, underscoring the potential of AI to foster deeper customer loyalty and streamline discovery.
Beyond technology teams, ThredUp sees potential for AI to revolutionize decision-making across the business. From evaluating A/B tests to assisting analysts, AI could drive a new era of data-informed workflows.
"If you're not embracing AI, it'll be harder to rise to the top of the talent pool," DeMeyere said.
Still, none of ThredUp's AI advancements would be possible without robust data practices. A single source of truth is important, he said.
"If customer behavior or lifetime value data isn't centralized, teams may draw conflicting conclusions," DeMeyere said.
ThredUp's data lakehouse aggregates inputs from operations, inventory and acquisition channels, creating a unified system that feeds into AI models.
"Dynamic models require constant oversight to ensure they align with intended goals," he said. This proactive monitoring ensures AI models adapt effectively to shifting data patterns.
