AI Pulse by Inblix

Nubank taps GPT-4o to handle 50% of support chats for 114M users

OpenAI Blog · Jul 14, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: Nubank taps GPT-4o to handle 50% of support chats for 114M users

Nubank is scaling its AI strategy in a way most fintechs only talk about. The Brazilian digital banking giant, now serving over 114 million customers across Latin America, has deployed a suite of OpenAI-powered tools that go well beyond a simple chatbot. They’ve built a call center copilot, an internal enterprise search tool, and a customer-facing assistant — all running on GPT-4o and GPT-4o mini — that are measurably changing how the company operates.

The numbers tell a compelling story. The AI assistant now resolves 55% of Tier 1 inquiries without any human intervention, handling over 2 million chat sessions and emails each month. That shift alone has slashed chat response times by 70%. On the agent side, more than 45% of Nubank’s support staff now lean on a copilot that serves up next-reply suggestions, summarizes long chat histories, and provides step-by-step guidance for complex issues. Queries are getting resolved 2.3 times faster, and the bank says its Transactional Net Promoter Scores are holding strong.

This didn’t start with customer support, though. Nubank’s first OpenAI project was decidedly internal: a custom enterprise search tool that uses Retrieval-Augmented Generation and fine-tuned models to let employees instantly pull answers from FAQs, policies, and brand guidelines. Over 5,000 employees now use it monthly, and the company credits it with faster onboarding and smarter decisions across developers and new hires alike. That foundation made the leap to customer-facing AI feel less like a gamble and more like a logical next step.

Still, the most quietly ambitious piece of this rollout might be in fraud. Nubank is piloting GPT-4o’s vision capabilities to analyze transaction records and submitted documents, combining text and image recognition to spot patterns that could indicate fraud. It’s a use case that demands precision, and the company stresses it’s done under strict policy and human oversight. If this works at scale, it won’t just be a win for Nubank — it’ll set a benchmark other digital banks will scramble to match.

💡 Key Takeaways

  1. Nubank’s AI assistant now resolves 55% of Tier 1 support queries alone, cutting chat response time by 70% across 2 million monthly interactions.
  2. Over 45% of human agents actively use a copilot that suggests replies and summarizes chats, helping resolve issues 2.3x faster.
  3. The company’s AI adoption began internally, with 5,000 employees using a GPT-4o-powered search tool before any customer-facing features launched.
  4. A fraud detection pilot uses GPT-4o vision to analyze both text and image data from transactions, hinting at far broader applications for multimodal models in banking.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles