Criteo introduced its Agentic Commerce Recommendation Service, designed to power AI shopping assistants with accurate, relevant product recommendations built on Criteo's commerce intelligence. LM platforms are rapidly evolving into AI shopping assistants, while retailers develop their own AI chatbots,encing how consumers discover, compare, and purchase products online. As these AI-driven shopping experiences scale, AI assistants need a commerce-grade recommendation infrastructure that drives outcome-based relevancy by accessing real shopping behavior, not just publicly available product descriptions, to deliver the trusted and personalized results that consumers expect.
This approach builds on Criteo's previously published agentic commerce vision. Built on Criteo's Commerce intelligence, its Agentic Commerce Recommendation service delivered up to a 60% improvement in recommendation relevancy compared to third-party approaches based only on product descriptions in Criteo's testing 1. This performance is enabled by the company's unmatched scale of 720 million daily shoppers, $1T in annual transactions, and 4.5 billion product SKUs. The service is available through Criteo's Model Context Protocol (MCP) and directly connects AI-powered shopping assistants with merchant inventory, translating consumer shopping queries into curated, transaction-ready product recommendations.
It enables AI assistants to surface the most relevant products for each individual consumer by applying real-world shopping and purchase signals that cannot be accessed through traditional c sprawling tactics. AI assistant query: The AI assistant queries Criteo's Agentic Commerce Recommendation Service to identify relevant products. Commerce intelligence-powered filtering: Criteo applies real-world shopping and purchase signal to filter and rank products based on what is most relevant for that individual consumer, considering nuances such as product popularity, availability, and user intent.
Curated results: Criteo returns a curated shortlist of product recommendations, rather than raw catalog data.


















