# Product Genius AI > Intent-based personalization for ecommerce. Product Genius is an AI commerce engine built on LIMs — Large Interaction Models — that continuously learn from every shopper interaction to optimize web and ad performance in real time. Delivered 36% average revenue lift in A/B tests announced with Google at NRF 2026; platform serves 20M+ shoppers and 840M+ interactions today. ## The problem Ecommerce teams are buried under pressure to improve performance faster than their tools, teams, and agencies can keep up. Markets evolve daily, audiences shift, and the cycle of content creation, curation, and testing moves too slowly to sustain growth. Results lag, insights arrive late, and consistent revenue lift remains out of reach. Agentic personalization ("TikTok-ifying ecommerce") is top of mind, but ecommerce personalization is fundamentally harder than social media's. Social platforms show a small catalog of trending content to users who scroll for hours daily, whose interests drift slowly and who generate dense behavioral data. Ecommerce faces the opposite on every axis: customers who visit infrequently and browse briefly, can't articulate what they want in advance, shift focus dramatically between — and within — sessions, and need instant access to an enormous catalog rather than a pre-curated feed. ## What Product Genius is Product Genius was built for exactly this problem. We've created **LIMs (Large Interaction Models)**, a next-generation AI architecture beyond LLMs. A LIM starts out trained like an LLM — predicting the next word — and then switches its objective to optimizing the next interaction with a shopper. It delivers **intent-based personalization**: continuously sensing, reading, interacting, and re-training in real time across multi-modal inputs — implicit behavioral signals, text, voice, and search — powering every shopper touchpoint from endless scrolls to ads. The underlying AI was developed with over $30M in DARPA funding and applied to ecommerce. ## Why prior personalization stacks fall short - **Legacy collaborative filtering** is session-blind, trapped in batch loops, serving "ghost" recommendations based on past purchases that are often no longer relevant. - **Hybrid search-ranking models** glue keyword indices to behavioral logs, require high-maintenance feature stores and manual rules, and still miss micro-intent. - **LLMs** finally reached the accuracy and reasoning to meaningfully interact with each shopper — but inference and training are a combination of too slow, too expensive, and too context-starved to sit behind every real-time interaction. None of these provide the scale, immediacy, and simplicity required for intent-based personalization across the many surfaces in a shopper's journey. ## How it delivers - **Speed** — learns and adjusts dynamically instead of on an A/B test schedule. Weeks become hours. - **Consistency** — compounding revenue lift as markets shift, not one-off wins. - **Control** — merchants own the growth loop; prompt the website and ads directly, no waiting on decks or dev queues. - **Differentiation** — maintains each brand's unique look and feel while adapting messaging, creative, and experience for each segment. - **Compounding network value** — every merchant benefits from a learning network of 20M+ shoppers and 840M+ interactions. Under the hood: - Learns new policies in as few as **3 interactions**, where standard methods require ~30,000. - Comprehends an entire catalog and shopper journey **50× more accurately than RAG**. - Scales independently of shopper volume and catalog size. ## Proof Announced with Google at **NRF 2026**: in controlled A/B tests across **17M+ shoppers and 700M+ interactions** on Shopify stores, Product Genius delivered **36% average revenue lift**, ranging from **12% to 130%** across hundreds of product verticals. Platform scale today: **20M+ shoppers and 840M+ interactions**. Product Genius powers TikTok-style endless scrolls and traditional recommendations experiences today, with email personalization and AdTech personalization in alpha — the same core engine delivering a consistent experience across any surface. ## Team Founded by pioneers at the intersection of AI infrastructure and neural architecture. Founder **Ben Vigoda** invented the original tensor-based deep learning processor design (exited to Analog Devices) and built first-to-market attention-based language models. Lead investor **Steve Papa** founded Endeca (became Oracle Commerce Cloud). Team experience includes scaling OfficeDepot.com to a $5B P&L and incubating Toast. ## Links - [Home](https://www.productgenius.ai/): Product overview and positioning. - [About](https://www.productgenius.ai/about): Company, mission, team. - [Demo](https://www.productgenius.ai/demo): Request a live walkthrough. - [Get started](https://www.productgenius.ai/get-started): Onboarding for new merchants. ## Deep dives - [Real-time personalization](https://www.productgenius.ai/real-time-personalization): How Product Genius personalizes the shopper experience on every page view. - [Stop firefighting](https://www.productgenius.ai/stop-firefighting): Moving from reactive revenue fire-drills to proactive AI-driven curation. - [Get back time](https://www.productgenius.ai/get-back-time): How merchandising automation returns hours to ecommerce teams. - [Twenty years](https://www.productgenius.ai/twenty-years): The research lineage behind the product. ## Content - [Blog](https://www.productgenius.ai/blog): Articles on ecommerce AI, personalization, and merchandising. - [News](https://www.productgenius.ai/news): Announcements and press. - [Careers](https://www.productgenius.ai/careers): Open roles. ## Legal - [Privacy policy](https://www.productgenius.ai/privacy-policy) - [Terms of service](https://www.productgenius.ai/terms-of-service) - [Cookie notice](https://www.productgenius.ai/cookie-notice) - [Data processing agreement](https://www.productgenius.ai/data-processing-agreement)