AI in banking is moving from simple chatbot interactions toward more contextual, multilingual, and accountable support. Customers increasingly expect an assistant that can help them find information, understand an issue, and take the next step without having to navigate a complex app or explain the same problem repeatedly.
bunq’s Finn illustrates that shift. It began as a conversational alternative to in-app search and has developed into a multi-agent financial assistant designed to resolve routine needs quickly while giving customers a clear route to human expertise when complexity demands it.
Read More
How Gupshup Is Making Enterprise AI Orchestration the New CX Control Plane
How bunq’s Finn Evolved From AI Search to a Multilingual Financial Assistant
For CX leaders, Finn’s development offers a useful example of where banking AI is heading. The question is no longer whether AI can answer a basic query. It is whether it can connect the right information, language support, specialist capability, and human oversight into one coherent customer journey.
TL;DR
- bunq launched Finn as a conversational way for customers to find transactions, understand spending, and navigate banking features in everyday language.
- Finn has expanded into multilingual AI customer support, with bunq reporting 38-language support, real-time translation, and a route to human specialists.
- Its multi-agent architecture shows how banking AI can scale specialist support while keeping trust, explainability, and customer control at the center.
Why Does bunq Finn’s Innovation Arc Matter for Financial Services CX?
The bunq Finn AI assistant matters because it reflects a broader change in banking CX: AI is shifting from isolated experimentation toward customer-facing systems that need to be useful, accountable, and easy to understand. Finn’s progression from natural-language search to AI customer support and financial guidance gives CX leaders a practical example of that transition.
In December 2023, bunq launched Finn as a generative AI platform that effectively replaced the search function in its app. The initial proposition was straightforward: users could ask questions about spending, saving, transactions, and app functionality in everyday language. Queries such as “What is the average amount I spend on groceries per month?” or “How much did I spend on Amazon this year?” made banking information easier to retrieve without navigating conventional menus.
Those early examples are important. Finn did not start as a narrow ticket-deflection tool. It started with a customer’s need to locate, interpret, and act on personal financial information. That foundation created a natural route toward the broader financial AI assistant role bunq describes today.
Ali Niknam, Founder and CEO at bunq, said:
“Years of AI innovation, coupled with laser focus on our users, allowed us to completely transform banking as you know it. Seeing Generative AI make life so much easier for our users is incredibly exciting.”
Key Takeaways: Finn’s Starting Point
- Finn began as a conversational way to retrieve and understand banking information.
- That customer-utility focus distinguishes the product from a purely cost-led automation strategy.
- The next stage of banking AI assistant design is about turning useful answers into coherent customer journeys.
The World Economic Forum’s 2026 AI Playbook for Financial Services, developed with Accenture, places that journey in a wider market context. Based on 18 months of discussions with more than 150 senior leaders across over 100 organizations, the report argues that financial institutions are moving beyond pilots. Their challenge is no longer simply gaining access to AI capabilities. It is building the data, governance, technology, and workforce foundations to use them securely at scale.
How Has the bunq Finn AI Assistant Expanded Since 2023?
The bunq Finn AI assistant has expanded from conversational banking search into a broader service and insight layer. bunq’s December 2025 update said Finn could provide more accurate answers and more human-like conversations, support users in 38 languages, and assist with questions spanning cards, payments, savings, merchant information, and travel recommendations.
bunq also reported that Finn had answered millions of user queries and achieved a 90% user satisfaction rating. The company said Finn resolves support queries in an average of 47 seconds, compared with the one-minute industry benchmark referenced in its announcement, and directs requests requiring deeper knowledge to the appropriate human specialist.
Moving From Information Retrieval to AI Customer Support
The difference between a search product and AI customer support is not simply the number of questions an assistant can answer. It is whether the system can understand a customer’s intent, access relevant information, and move the interaction toward resolution. A question about a failed payment, for example, may require an explanation, a transaction check, product guidance, and potentially a specialist handoff.
According to bunq, Finn handles around 97% of user support activity. AWS’s January 2026 technical case study also reported 97% coverage and a 47-second average response time. Its automation figures vary within the article, with one section referring to more than 82% fully automated and an attributed quote citing 70%. Those figures should not be treated as directly comparable without matching definitions or reporting periods. The more reliable editorial conclusion is that bunq and AWS both describe high-volume automation coupled with human escalation.
Why Multilingual Customer Service Is Part of the Experience
Finn’s 38-language capability turns multilingual customer service into more than a localization feature. AWS says Finn can translate the bunq app and deliver real-time speech-to-speech translation during conversations with the support team. For customers who travel, work internationally, or manage money across borders, that can reduce the effort of explaining an urgent issue in a non-native language.
However, language coverage alone does not make a service journey trustworthy. The experience also needs to preserve meaning, communicate what will happen next, and make it clear when a customer is being transferred to a person. That is where product design and operational governance meet.
Key Takeaways: Scaling Customer Support
- bunq says Finn supports 38 languages, including real-time speech-to-speech translation.
- Reported 47-second response times are meaningful only when answers are relevant, safe, and understandable.
- Human escalation keeps AI customer support connected to specialist expertise.
For CX leaders, the lesson is to measure a banking AI assistant beyond containment. Speed and automation matter, but they are not the full customer outcome. A successful service interaction should leave the customer with a clear answer, a realistic next action, or a frictionless path to someone who can help.
Why Did bunq Move to a Multi-Agent Architecture?
bunq moved to a multi-agent architecture to make its AI customer support system more adaptable as its service requirements grew. AWS explains that an initial router-based model became harder to manage because more specialist agents created routing complexity, overlapping capabilities, and a potential bottleneck whenever a new agent or capability was added.
The revised design uses an orchestrator that routes a request to a limited number of primary agents. Those agents can call specialist agents as tools when needed. Rather than requiring a central router to predict every possible customer need, the system lets primary agents request specific capabilities, such as transaction analysis, knowledge retrieval, or account support, during the interaction.
What Does Agent Orchestration Mean for the Customer?
For the customer, agent orchestration should mean less visible complexity. A failed-payment query may require different forms of expertise, but the user should not need to understand bunq’s internal systems or repeat their explanation across handoffs. The value of the architecture is its ability to coordinate relevant steps behind a single, coherent interface.
AWS says Finn uses Anthropic Claude models through Amazon Bedrock, alongside Amazon ECS, DynamoDB, OpenSearch Serverless, S3, and MemoryDB. In practical terms, the stack supports model access, deployment, context retention, semantic retrieval, document storage, and session management. Amazon Bedrock is therefore an important part of the solution, but not the entire customer experience.