Shopify Deploys New Generative Recommender

Shopify's new recommender system is framed as a "foundational generative recommender," designed to predict a user's next action by interpreting their entire history as a sequence. This moves beyond traditional collaborative filtering, which often summarizes user behavior, by analyzing the specific order of searches, clicks, add-to-carts, and purchases over months. The system was co-developed with AI startup Liquid AI and leverages their proprietary HSTU (Hierarchical State-Space Transformer Unit) architecture. In controlled testing, this new model resulted in higher conversion rates compared to Shopify's previous recommendation stack. This initiative is part of a multi-year partnership with Liquid AI, in which Shopify also participated in a $250 million Series A funding round in December 2024. The collaboration has already produced a new AI-powered search model for storefronts that delivers results in under 20 milliseconds. The engineering team, including Senior Staff Machine Learning Engineers Yang Liu and Ali Khanafer, focused on treating time as a critical signal within the model. They implemented a novel time-encoding method so the system understands that the timing and gaps between events carry meaning, making recommendations more context-aware. A key challenge at Shopify's scale—which saw 81 million consumers during the 2025 Black Friday Cyber Monday weekend—is adding incremental value. The model is trained to find products that other, existing recommendation models might miss, ensuring it provides new and genuinely helpful suggestions to shoppers. This generative approach is part of a broader strategy at Shopify to embed AI into core commerce functions. This includes "Sidekick," an AI assistant for merchants, and the Universal Commerce Protocol, co-developed with Google, to standardize transactions initiated by AI agents on platforms like ChatGPT and Gemini.