Nubank
genAI customer agent + agent copilot
In 2024, Nubank's customer assistant built with OpenAI (GPT-4o) handles over 2 million chats per month and resolves 55% of Tier 1 inquiries without a human, with chat response time reduced by 70% across a base of 114 million customers.
Key points
- genAI customer assistant and agent copilot for customer service.
- Built with OpenAI GPT-4o and in-house integration.
- Over 2 million chats per month, 55% of Tier 1 inquiries resolved without a human, -70% response time.
- Evidence B, confirmed status.
Objective
Absorb the customer service volume of a base of over 100 million customers without blowing up costs, by automating Tier 1 inquiries and equipping human agents with a copilot to move faster on the rest.
The deployment
Nubank deployed, with OpenAI on GPT-4o, two conversational components. A customer assistant holds over 2 million chats per month and resolves around 55% of Tier 1 inquiries without going through a human, with chat response time reduced by 70%. In parallel, a call center copilot assists agents in real time (conversation summaries, sentiment analysis, suggested responses) and is used by over 45% of agents. The bank serves over 114 million customers in Brazil, Mexico, and Colombia.
Results Proof B
Figures from the OpenAI customer story (T2, official but interested source) and picked up by the fintech press (T4), concordant. Operational metrics provided by the platform and the brand, not audited in financial results, hence B.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- llm OpenAI GPT-4o Model behind the customer assistant and the agent copilot, including vision for document analysis.
- outil Assistant client in-house Conversational component integrated into the Nubank app: over 2 million chats per month, 55% of Tier 1 inquiries resolved without a human.
- outil Copilote de centre d'appels Conversation summaries, sentiment analysis, and suggested responses in real time on the agent's desk; used by over 45% of agents.
- integrateur OpenAI (partenariat) Partner in the deployment of both conversational components.
How it runs, concretely
For ops teams-
1Incoming chat AI
The customer opens a chat; the assistant handles Tier 1 inquiries (account, card, payment) and closes what it can.
-
2Escalation to an agent AI / customer service
Unresolved inquiries pass to a human agent with the conversation context.
-
3Agent copilot AI / customer service
On the agent side, the copilot provides a summary, sentiment analysis, and suggested responses in real time.
-
4Monitoring and tuning data team
The data team tracks resolution, response time, and copilot adoption to broaden the scope handed to AI.
The Tier 1 resolution rate without escalation and customer sentiment. If Tier 1 resolution drops or sentiment degrades, volume flows back to agents and the productivity gain disappears.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- Support ticket history labeled by tier
- Real-time access to the customer account
- Call transcripts to train the copilot
Org prerequisites
- An assistant-to-agent handoff rule
- Copilot adoption on the agent side (change management)
- A compliance framework for using a third-party LLM on customer data
Possible stack
- Generative LLM (OpenAI GPT-4o or equivalent)
- RAG layer over a knowledge base
- Integration with ticketing and the agent's desk
The plan, step by step
- Step 1Identify the highest-volume Tier 1 inquiries in the ticket history and define a clear escalation rule to a human.Deliverable: Inquiry taxonomy and validated handoff rules.
- Step 2Build the assistant on the LLM with access to account data, in a test environment, with a compliance framework for using a third-party LLM.Deliverable: Assistant in pre-production on 3 to 5 inquiry families.
- Step 3Launch a production pilot on a fraction of chat traffic and measure Tier 1 resolution and response time.Deliverable: Pilot readout with the resolution-without-a-human rate.
- Step 4Broaden the scope of inquiries covered and deploy across all Tier 1 traffic.Deliverable: Generalized assistant with a resolution/satisfaction dashboard.
- Step 5Launch the agent copilot (summaries, sentiment, suggestions) and drive adoption with the teams.Deliverable: Active copilot with agent adoption rate tracked.
First step: First automate the highest-volume Tier 1 inquiries with a clear escalation rule, then equip agents with the copilot.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Nubank elevates customer experiences with OpenAI Interested party archive pending
- S2 Nubank Elevates Customer Experiences with OpenAI Secondary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.