AI in Brazil

Brazilian plant deployed AI-driven predictive maintenance across compressors, HVAC systems, and packaging equipment, achieving $2.3M annual savings on a $5.1M maintenance baseline, 40 percent reduction in unplanned downtime, and OEE above 85 percent for two consecutive years, the highest in Unilever's global manufacturing network.

In a side-by-side trial on Maristela Storti's corn farm in Jatai, Goias, Stenon's FarmLab soil sensor and WebApp produced variable-rate nitrogen recommendations that used 20% less ammonium sulfate than the farm's standard rate and gave 7.98% higher yield, adding US$163.61 profit per hectare. Stenon projects over US$347,673 if applied across the farm's 2,800 hectares.

Suzano, the world's largest pulp producer, built VagaLúmen with Google Cloud partner Sauter Digital: Cortex Framework populates views of SAP materials master data in BigQuery, and Gemini Pro 1.0 translates employees' natural-language questions into SQL. Since going live in May 2024 it has had 4,000 accesses from 473 unique users, and querying input and MRO materials fell from about 2 minutes to 8 seconds.

Magalu built Lu's Brain, an AI shopping agent on the WhatsApp Business Platform hosted by its virtual persona Lu, that handles search by text, voice or image, recommends products in card carousels, shows product pages via WhatsApp Flows and takes Pix payments. In a six-week 2025 pilot it reached 3X the conversion of Magalu's app and site and an 85 NPS.

iFood built ML Go!, an internal ML platform on Amazon SageMaker that standardizes training and deployment with SageMaker Pipelines, Model Registry, Inference Components and its BruceML scaffolding tool. It serves recommendations, delivery-time prediction, a customer-service chatbot and a grocery assistant, and iFood helped AWS develop features that cut inference costs.

iFood's AI Academy built a route simulator on AWS that tests logistics parameters by weekday and time of day in a sandbox before releasing models to production. iFood reports a 12% cut in courier travel distance, 50% less courier idle time and on-time delivery rising from 80% to 95%, across 39 million monthly orders.

Nubank built four tools with OpenAI: a GPT-4o enterprise search used by 5,000+ employees a month, a Call Center Copilot used by over 45% of agents, a customer AI Assistant handling 2M+ chats a month that resolves 55% of Tier 1 inquiries and cut chat response time 70%, and a GPT-4o vision pilot for fraud checks.

Nubank built nuFormer, a proprietary AI model for analyzing customer behavior, and used it to adjust credit card limit policies in Brazil so it could extend more credit to qualifying customers without changing its overall risk appetite. CEO David Vélez said the bank also sees room for agentic workflows across most products.