Transforming Banking: How bunq’s AI Assistant Finn Revolutionizes Customer Support with Amazon Bedrock
Revolutionizing Banking with Agentic AI: The Case of bunq’s Finn
In an age where automation is vital, bunq, Europe’s second-largest neobank, is at the forefront of a revolution. Co-authored by Benjamin Kleppe, Machine Learning Engineering Lead at bunq, we delve into how agentic AI is reshaping the banking industry and transforming customer support systems.
The Rise of Agentic AI in Banking
The integration of agentic AI signifies a monumental shift from conventional customer service models. Unlike traditional support systems, which often find it challenging to manage complex financial products and meet the growing demands for immediate, accurate responses, agentic AI thrives on autonomy. It empowers banks to provide 24/7 multilingual assistance, process transactions swiftly, and deliver bespoke financial insights on a significant scale.
bunq: Putting Users First
Founded in 2012 by entrepreneur Ali Niknam, bunq’s mission is simple: to enhance the banking experience for those with an international lifestyle. With 20 million users across Europe, bunq focuses on ease of use, making spending, saving, budgeting, and investing seamless—all within a single application designed from user feedback.
Yet, with user-centricity comes responsibility. Bunq faced significant challenges in delivering consistent, high-quality support across various channels, languages, and time zones. Traditional methods couldn’t keep pace with the demand for instant assistance, thus necessitating a more sophisticated solution.
The Challenge
Consumers expect immediate access to banking services, whether it’s resolving transaction disputes or seeking financial advice. For bunq, existing systems created bottlenecks, leading to unsatisfactory user experiences. Moreover, they required a robust means of analyzing user feedback to continuously evolve their offerings.
Enter Finn: A Revolutionary Solution
In 2023, bunq rolled out Finn, an innovative generative AI assistant built entirely in-house. Finn utilizes leading AI foundation models (FMs) and technologies, including Anthropic’s Claude models via Amazon Bedrock. Unlike generic chatbots, Finn isn’t just reactive; it processes natural language and provides intelligent, real-time answers.
Finn’s unique capabilities include:
- Multilingual Support: Translating the bunq application into 38 languages.
- Real-time Speech Recognition: Converting speech-to-speech calls to the support team instantly.
- Task Automation: Handling processes such as invoice recognition with remarkable efficiency.
- Financial Insights: Offering personalized budgeting advice and complex banking information summaries.
Implementing a Smarter AI Solution
The deployment of Finn utilized Amazon Web Services (AWS), ensuring a secure, scalable, and compliant infrastructure. Here’s how:
- Amazon Bedrock: A fully managed service that provides robust FMs and enables bunq to leverage Anthropic’s Claude models securely.
- Amazon Elastic Container Service (ECS): Facilitates the management of containerized applications, allowing the development team to focus on creating Finn’s multi-agent architecture.
- Amazon DynamoDB: Powers high-performance applications, managing user memories and conversation histories to maintain context.
- Amazon OpenSearch Serverless: Supports semantic searches across bunq’s knowledge base, enabling efficient information retrieval.
Rethinking Architecture: From Routers to Orchestrators
Initially, bunq’s system operated on a router-based model directing queries to specialized agents. However, they encountered challenges like routing complexity and scalability bottlenecks. The solution? A re-engineered architecture centered around an orchestrator agent.
This orchestrator simplifies query routing, directing users to a handful of primary agents who can dynamically invoke specialized agents as needed. This flexibility allows firmer responses to user inquiries without the overhead of predicting every possible scenario.
Real-World Results and Impact
With this new architecture, Finn achieved remarkable operational efficiency:
- 97% of Support Activity Managed: Finn now handles almost all customer support interactions, with over 82% automated.
- Rapid Response Times: Average response times have decreased to just 47 seconds.
- Accessibility: By supporting 38 languages, bunq has made banking accessible to millions.
Benjamin Kleppe highlights this evolution: “We went from concept to production in just three months. Now, Finn processes 97% of inquiries with 70% automated.”
Conclusion: A New Era for Banking
bunq’s journey illustrates the profound impact of adopting agentic AI. Through innovative architecture and a user-centric approach, bunq is reshaping the banking landscape and setting a new standard for customer service.
With Finn leading the charge, bunq has positioned itself as Europe’s first AI-powered bank, delivering capabilities that far surpass those of traditional support systems. As we move forward, it’s clear that the future of banking lies in intelligent AI solutions that allow institutions to focus on what truly matters—enhancing the user experience.
For those interested in leveraging AI power within banking, exploring tools like Amazon Bedrock and foundation models can lead to transformative outcomes.
About the Authors
Benjamin Kleppe is the Machine Learning Engineering Lead at bunq, supervising the development of intelligent banking solutions. He focuses on enhancing user experiences and automating complex banking processes.
Jagdeep Singh Soni is a Senior AI/ML Solutions Architect at AWS, specializing in generative AI and Amazon Bedrock. His mission is to help organizations build innovative AI applications.
Guy Kfir is the generative AI Lead at AWS, assisting enterprises in adopting AI solutions and shaping go-to-market strategies.
By navigating this landscape together, banking and technology can create a seamless, efficient, and user-friendly experience that meets the demands of today and tomorrow.