• Every day, shoppers from Etsy’s community of nearly 90M buyers visit our marketplace to search for unique, handmade, and vintage items. • But with over 100 million listings, how do we help each buyer find exactly what they’re looking for? • Traditional search and recommendation systems often fall short of capturing the nuanced interests that make each Etsy buyer unique, from specific styles to aesthetic preferences. • At Etsy, understanding our buyers’ interests is central to delivering engaging, personalized experiences. • Recently, we explored enhancing our personalization by leveraging large language models (LLMs) to create detailed buyer profiles based on buyers’ browsing and purchasing behaviors. • We strive towards privacy by design and build these exploratory models with buyer privacy in mind.

Article Summaries:

  • Etsy is testing a new personalization approach that uses large language models (LLMs) to generate anonymous buyer profiles from browsing and purchase data. The system pulls recent searches, views, and purchases, then prompts an LLM to infer interests such as preferred styles, product categories, and shopping goals. To keep the process scalable and cost‑effective, Etsy shifted to BigQuery for faster data retrieval, trimmed the input window to the last nine months, increased batch sizes, and introduced parallel processing. These optimizations cut profile‑generation time from 21 days to three days for ten million users and lowered the estimated cost by 94 % per million users, while maintaining privacy‑by‑design safeguards.
  • Etsy is testing large language models (LLMs) to generate detailed buyer profiles from users’ browsing and purchase histories. By pulling activity data from internal feature stores and BigQuery, the company prompts an LLM to infer interests such as preferred styles, product categories, and shopping missions, while preserving privacy. To make the approach scalable, Etsy shifted to clustered BigQuery tables, trimmed input to the last nine months, increased batch sizes, and added parallel processing. These optimizations cut profile‑generation time from 21 days to three days for 10 million users and reduced the cost per million users by 94 %. The initiative remains experimental but demonstrates a path toward large‑scale, cost‑effective personalization.
  • Etsy is testing large‑language models (LLMs) to generate detailed, privacy‑preserving buyer profiles from browsing and purchase data. By pulling recent activity from internal feature stores and BigQuery, the LLM interprets sessions to identify style preferences, product categories, and shopping goals, then outputs structured profiles with confidence scores. To scale to 90 million users, Etsy shifted to clustered BigQuery tables, trimmed input data to the last nine months, increased batch sizes, and introduced parallel processing. These optimisations cut profile‑generation time from 21 to 3 days for 10 million users and lowered per‑million‑user costs by 94 %, making large‑scale personalization economically viable.
  • Etsy is experimenting with large language models (LLMs) to generate detailed, privacy‑preserving buyer profiles from users’ browsing and purchase histories. By ingesting recent search, view, and purchase data, the LLM produces structured profiles that capture style preferences, product categories, and shopping goals. To scale the approach to its 90 million‑user base, Etsy shifted to BigQuery for efficient data access, trimmed input data to the last nine months, increased batch sizes, and introduced parallel processing. These optimizations cut profile‑generation time from 21 days to 3 days for 10 million users and reduced per‑million‑user costs by 94 %. The initiative remains in the experimental phase.
  • Etsy is experimenting with large language models (LLMs) to build detailed, privacy‑preserving buyer profiles from users’ browsing and purchase histories. By pulling recent activity from internal feature stores and BigQuery, the LLM interprets interactions to generate structured profiles that capture style preferences, category interests, and shopping goals. To scale the approach, Etsy shifted to clustered BigQuery tables, trimmed input data to the last nine months, increased batch sizes, and introduced parallel processing, cutting profile‑generation time from 21 to 3 days for 10 million users. Cost reductions were achieved by using smaller models and optimized prompts, lowering the estimated expense by 94 % per million users. The project remains in experimental stages but demonstrates a path toward large‑scale, personalized recommendations.

Sources: