Customer Service Automation · Enterprise Knowledge Search · Fraud Detection
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.
Reported impact
Nubank accelerated employee search, handled millions of customer conversations with automation, cut chat response time by 70%, resolved queries 2.3x faster, and added vision-based fraud QA to improve consistency and service quality.
Nubank built a connected AI support stack with OpenAI that starts inside the company and extends directly to customers. The first layer is a custom enterprise search tool that uses GPT-4o, GPT-4o mini, RAG, semantic search, and fine-tuned models to retrieve FAQs, brand guidelines, and policies from fragmented internal repositories through a chat interface. On top of that knowledge foundation, Nubank created a support copilot with GPT-4o that surfaces next-reply suggestions, chat summaries, and step-by-step guidance for agents around the clock. The bank then added an AI Assistant powered by GPT-4o that handles up to five automated interactions directly with customers before escalating, and separately began piloting GPT-4o vision for fraud-quality assurance across transactions, documents, and customer communications. This layered design lets Nubank reuse knowledge retrieval, context, and supervision patterns across employee productivity, support automation, and fraud operations.
Internal documents are retrieved through RAG and semantic search, then ranked and answered through the enterprise search interface. The same knowledge and chat context power the agent copilot and AI Assistant for customer support, while GPT-4o vision separately reviews transaction and document evidence to flag fraud issues under policy controls.
"AI-powered assistant: Resolves 55% of Tier 1 inquiries, handling over 2 million monthly chats and reducing chat response times by 70%."
"Nubank is piloting a solution to ensure more consistent and high-quality responses to fraud incidents with GPT-4o vision."
"Powered by GPT-4o and GPT-4o mini, the solution integrates Retrieval-Augmented Generation (RAG) techniques"
"over 5,000 employees use it monthly"
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
Microsoft Sample Jan 8, 2025 Aditya Birla Capital It wanted to streamline the business and deliver a more connected customer experience using generative AI. Financial services provider Aditya Birla Capital has a massive market presence with more than 40 million… Azure AI Search · Azure AI Services · Azure Cosmos DB Microsoft Sample Jan 11, 2024 MONETA Money Bank Customer center MONETA Money Bank built by NTT Czech Republic uses an AI-driven voicebot, powered by Feedyou Platform and Microsoft Azure AI, to converse fluently in the complex Czech language, and can gather customer… As a result, the bank cut call center operational costs by 10 percent, while increasing customer satisfaction. Azure AI Services · Azure AI Speech · Azure Machine Learning 
OpenAI Sample Jul 7, 2026 MUFG MUFG needed to modernize banking operations and customer experiences with AI while satisfying the governance, security, training, and trust requirements expected of a major financial institution. 100% training participation among employees receiving accounts AI bankers · ChatGPT Enterprise · MAP (Money Advisory Platform) 