Customer Service Automation · Fraud Detection
Mercado Libre needed an AI platform that could safely automate complex customer-service mediation and other high-stakes workflows across a massive commerce and fintech environment, not just assist with isolated catalog or review tasks.
Reported impact
Mercado Libre expanded catalog coverage, improved fraud accuracy, increased order conversion, and began shifting large volumes of customer-service mediation from human operators to an internal AI platform built for secure scale.
Mercado Libre built Verdi as a natural-language-first development platform that lets its 17,000 developers create AI applications across more than 30,000 microservices while keeping security, routing logic, and guardrails built into the platform. Verdi uses GPT-4o, GPT-4o mini, and GPT-3.5 Turbo to handle multimodal reasoning, routing, and lightweight tasks, and one of its first major deployments is autonomous customer-service mediation between buyers and sellers. Rather than forcing developers to manage code-level complexity directly, Verdi presents a structured framework of language models, Python nodes, skills, and APIs so teams can compose agentic workflows, including image-based “proof of damage” steps with vision, inter-skill communication, unit testing, and deployment inside Mercado Libre’s internal developer platform. Beyond Verdi, OpenAI models also power inventory expansion, fraud detection, review summarization, product-description localization, and personalized notifications across the broader marketplace.
Marketplace listings, disputes, reviews, images, and customer-service cases flow into Verdi’s skill graph, where GPT-4o, GPT-4o mini, and GPT-3.5 Turbo combine with Python nodes and APIs to make decisions, call systems, and return outputs. The same model family also enriches listings, detects fraud, summarizes feedback, and personalizes notifications across the marketplace.
"In just a few months, Verdi has started to handle 10% of customer service mediation"
"The biggest difference between Verdi and other solutions is that in Verdi, natural language remains the central mode of interaction"
"Verdi changes this by allowing the AI to manage these interactions"
"improving fraud detection accuracy to nearly 99% for flagged items"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Aug 14, 2026 Shopee Shopee needed large language models that could handle multilingual ecommerce tasks across Southeast Asia, Taiwan, and Brazil, but general-purpose frontier models degraded sharply on regional languages, shopping… 113x growth from 3 billion to 340 billion monthly tokens Compass v3.5 · NVIDIA A100 GPUs · NVIDIA H100 GPUs 
OpenAI Sample Mar 6, 2025 Nubank Nubank needed to scale employee knowledge access, customer support, and fraud-response quality across more than 114 million customers without slowing response times or overloading human agents. 70% reduction in chat response times AI Assistant · GPT-4o · GPT-4o mini OpenAI Sample Modified · Oct 16, 2026 Grupo Boticário Grupo Boticário needed to support a complex multi-brand reseller and franchise ecosystem across many channels while reducing repetitive operational work and preserving the human relationships that drive sales and… 100% of conversations generated in appchat and WhatsApp Enterprise covered since November 2025 OpenAI foundation models · WhatsApp Enterprise · appchat 