Meta Tests AI Shopping Features

This initiative is part of Meta's broader push into "conversational commerce," a market projected to reach over $39 billion by 2034. The goal is to collapse the traditional shopping funnel, moving users from discovery to purchase within a single chat interface and reducing friction that leads to abandoned carts. The technology likely leverages Meta's own Llama 3 large language model, which is designed to be multilingual, multimodal, and have a longer context window for more natural conversations. For product recommendations, Meta's systems use a multimodal ranking approach, combining its Lumos image understanding platform and DeepText engine to analyze both product images and descriptions for more accurate matching. This move intensifies competition with Pinterest's Lens and Google Lens, which also use visual search to drive e-commerce. While Google Lens has broader object recognition, Pinterest Lens is highly optimized for lifestyle, fashion, and home décor—key verticals for social commerce. Meta's advantage lies in integrating this capability within its massive social graph and messaging platforms. Underpinning these features is Meta's "Advantage+" suite of AI-powered advertising tools. Advantage+ Shopping campaigns, which automate ad processes and targeting, achieved a $10 billion run rate by 2023 and saw 70% year-over-year growth in Q4 of the same year, indicating strong advertiser adoption of AI-driven commerce solutions. For MLOps, deploying such a system at scale requires a robust infrastructure. Facebook's recommendation architecture often utilizes a multi-stage ranking system. A retrieval system first selects a shortlist of relevant items from billions of listings, and then a more computationally intensive ranking system, often a neural network, determines the final recommendations to display to the user. Meta is also experimenting with generative AI to create ad variations and even entire campaigns automatically, with a goal of full automation by the end of 2026. This includes tools for text variation, background generation, and image outcropping to fit different formats like Reels and Stories, which are tested in an environment called the AI Sandbox. Beyond chat, Meta is testing other AI-driven shopping features, including virtual try-on tools that allow users to upload their own photos to see how apparel might look. This leverages the company's Spark AR platform, which uses computer vision to overlay 3D objects onto real-world environments.