Grubhub, part of Wonder, is a leading U.S. ordering and delivery marketplace dedicated to connecting customers with their favorite local restaurants, merchants, and convenience retailers. Time to launch isn't just an operational metric. For every merchant on the platform, it's the difference between earning and waiting. To solve it, Grubhub partnered with Invisible to build an AI-powered pipeline that uses intelligent routing and AI agents to extract key menu information from unstructured submissions into structured, platform-ready data. The result is fewer errors and faster merchant activation.
Our merchants succeed when we remove friction from day one, and menu onboarding is where that starts. Partnering with Invisible, we built an AI-powered pipeline that handles the complexity of merchant menu onboarding automatically, at the speed and accuracy our merchants expect. Less manual work, fewer errors, and faster time to revenue for every restaurant on our platform.
Every new merchant on Grubhub needs a menu on the platform before they can take their first order. The problem: merchants submit menus in a variety of formats - handwritten notes, PDFs, website URLs, photos, and more.
Grubhub's team had to manually extract and reformat every menu item, price, and order modifications to meet the platform's precise data requirements, which made the process arduous and error-prone.
Every minute of delay was a day a merchant wasn't generating revenue.
Invisible built an end-to-end automated pipeline that moves thousands of menus monthly from intake to platform in seconds – a decrease of 75-80% from manual processes.
A real-time API integration with Grubhub's delivers menu data to Invisible the moment it’s ready. The pipeline then automatically classifies each incoming menu by format and routes it to one of two custom AI extraction tools: one built for URL-based menus, one for PDFs and images. Both tools are purpose-built for Grubhub's exact menu structure, trained to parse complex modifier hierarchies and output clean, structured data.
Every AI output goes through a human-in-the-loop review, where human experts make targeted corrections before the menu moves forward. The result is a finished, platform-ready menu that meets Grubhub's quality bar every time.
