Nubank's AI customer-service stack, built on OpenAI's GPT-4o, resolves more than 50% of Tier 1 inquiries, handles over 2 million monthly chats, and cuts chat response times by 70%, all while serving more than 114 million customers across Brazil, Mexico, and Colombia. "Together, Nubank and OpenAI developed a Call Center Copilot to assist agents in real time by integrating the bank's knowledge base and chat history," OpenAI writes in its customer case study, noting that the solution runs on GPT-4o's real-time multimodal text-and-speech capabilities — "essential for good customer support."
Customer support has been a chronic scaling problem at Nubank as the neobank grew from a Brazilian startup founded in 2013 into a multi-country platform. The deployment actually combines four OpenAI-backed pieces. First, an internal enterprise search powered by GPT-4o and GPT-4o mini with retrieval-augmented generation, used by over 5,000 employees monthly for FAQs, brand guidelines, and internal policies. Second, a Call Center Copilot that suggests next replies, summarizes chats, and walks agents through complex issues, with over 45% of Nubank agents using its key features. Third, a customer-facing AI Assistant, also on GPT-4o, that handles up to 5 automated interactions before escalating and processes more than 2 million chats monthly. Fourth, a fraud-quality-assurance pilot using GPT-4o vision to analyze transaction records and submitted documents.
End-to-end, Nubank reports resolving customer queries 2.3x faster while maintaining strong transactional Net Promoter Scores. The enterprise search additionally uses models fine-tuned on Nubank's domain content and semantic search to rank results, and the vision fraud pilot combines natural language processing with image recognition to spot anomalies in line with fraud-prevention policies, regulatory requirements and oversight by dedicated teams.
Details
- City
- São Paulo
- Organization
- Nubank
- Continent
- South America
- Country