September 21, 2026
iFood Trains AI on the Behavior of People Who Are Already Customers
iFood is putting R$ 2 billion into agents and the Large Commerce Model, trained on orders, cancellations and reviews. Operational data became raw material.
At iFood MOVE, iFood announced R$ 24 billion in investment through March 2027, 41% above the previous cycle, to grow beyond food delivery. Within that package, more than R$ 2 billion goes to AI agents and to the Large Commerce Model.
The Large Commerce Model is a generative AI model that iFood built with Prosus, trained on anonymized searches, purchases, cancellations and reviews to interpret purchase intent. In August, I wrote here about how iFood measured the results generated by its 27,000 AI agents and how the development team gained 16% in productivity with that metric.
The model learned from iFood's customer history and, according to Prosus, also from the group's transactions, which span more than 500 million users. According to iFood, the result is more precise notifications and recommendations, at a lower cost than equivalent proprietary models.
For anyone building products, the data an operation already generates, such as orders, cancellations and reviews, is raw material for a model that reads purchase intent.
Sources: Forbes Brasil · Correio Braziliense · O Tempo · Prosus