Our journey to scalable GenAI in banking

My Data Science journey

My career has always been about finding that sweet spot where curiosity intersects with real-world impact. With a background in mathematics, I’ve always been drawn to solving complex problems, optimizing systems, and understanding underlying patterns. Math gave me the tools to think abstractly, but data science gave me the ability to apply those ideas to real-world challenges.

I started out with a curiosity about applied science—finding practical ways to improve systems, processes, and experiences. This curiosity led me into the world of algorithms, programming, and eventually artificial intelligence. Over the years, I’ve worked on projects ranging from building neural networks from scratch to implementing advanced models like deep learning systems and reinforcement learning algorithms—all with a focus on performance and efficiency.

The magic of NLP and Generative AI

Natural Language Processing (NLP) has always intrigued me. I’ve watched it evolve from simple text-processing techniques to the game-changing generative AI of today, thanks to breakthroughs like the Attention Is All You Need paper. And let’s be real: laziness really is the father of innovation. The magic of AI lies in its ability to automate repetitive tasks, allowing us to focus on the work that truly adds value and drives impact.

This philosophy—optimizing workflows and transforming the mundane into something smarter—has not only shaped my career but also driven the journey of our team.

From Data Science to end-to-end solution builders

When we started, our team was dedicated to financial crime prevention, developing an alert prioritization model that would sharpen the bank’s defences against money laundering. This was crucial work with significant impact for both compliance and security, and our model was successfully brought to production. But as we continued to optimize, we began to see the potential for more—we wanted to tackle new challenges where we could drive value across more areas of the bank. We set our sights on creating something transformative, something that could reshape how we use AI at scale.

It was a pivotal moment for us. While we initially focused on data science, we quickly realized the potential to broaden our scope and build fully integrated solutions. To kick things off, we held a brainstorming session where everyone pitched their boldest ideas. The list grew rapidly, and before long, we had a wide range of ideas—from automating responses to alerts and building smarter customer profiles to developing AI tools for fraud detection, compliance, and cross-selling. With so many exciting possibilities, a few ideas quickly rose to the top: using GenAI to automate processes across the bank.

With that clear goal in sight, we dove headfirst into the world of generative AI.

Starting our GenAI journey against the odds

Our excitement was met with challenges. Strict bank policies and compliance concerns restricted access to popular generative AI tools like ChatGPT and Copilot. But we were determined. We reached out to a neighbouring team working on an approved GenAI project and suggested a collaboration. This collaboration became the foundation for our first official use case and the creation of the "GenAI Guild"—a shared space where we could exchange knowledge, code, and insights, equipping us to build compliant, scalable GenAI solutions in-house.

Expanding our full-stack capabilities

With this shared space, we grew from being data science experts into a team capable of handling the full stack of GenAI development. From the first summarization use case, we began expanding our skills in infrastructure, compliance, and user feedback—soon, we weren’t just exploring; we were creating.

Building and scaling GenAI solutions

One of our earliest breakthroughs came with a new use case: Retrieval-Augmented Generation (RAG). RAG was the perfect playground for us to test our end-to-end capabilities. Working collaboratively, our data engineers built the necessary infrastructure and feedback mechanisms to make our solution compliant, scalable, and seamless.

As our knowledge grew, we moved from using external GenAI libraries to developing our own robust solutions. This shift required a real transformation from traditional data science to a comprehensive approach, tackling every part of the pipeline. Through continuous experimentation, we optimized the retrieval quality and accuracy of our models, which established us as a leading GenAI team within the bank.

Tackling evaluation in the world of GenAI

Evaluation is central in data science—it’s how we refine and validate our models. But with GenAI, traditional evaluation methods didn’t always apply, so we took the challenge of designing new evaluation methods. We started building an evaluation package tailored to GenAI, and it soon expanded to help other teams evaluate their RAG solutions and add a red-teaming component to test for vulnerabilities.

This evaluation package proved extremely valuable, and we realized that this kind of end-to-end solution creation was exactly the shift we had been seeking. Not only did we improve the chatbot’s performance, but we also showed that automated evaluation of GenAI models was feasible, raising the bar for model monitoring and reliability.

Enter “GenAI Managed Services”: our solution to scaling GenAI across the bank

As our GenAI efforts grew, we recognized a need to streamline access to GenAI solutions across the bank. Initially, we worked closely with individual teams, but this approach wasn’t scalable. That’s when we introduced GenAI Managed Services.

GenAI Managed Services acts as a centralized 'Store' —a one-stop-hub for all things related to GenAI. It provides easy access to pre-built tools and services such as Retrieval-Augmented Generation (RAG) solutions, summarization tools, ans Responsible GenAI Toolkits. By streamlining access ans standardizing adoption, this approach has made it easier for teams to integrate GenAI into their workflows while maintaining cosistency and compliance.

Today, we’ve evolved from a specialized team focused on financial crime prevention into a GenAI powerhouse, organized into dedicated workstreams to build, refine, and scale GenAI services across the bank. With "GenAI Managed Services" up and running, we can now empower teams everywhere to build faster, more confidently, and with scalable, compliant GenAI solutions.

Our journey has been as much about ownership as it has been about innovation. As we expanded into full-stack development, we took responsibility for the entire lifecycle—from infrastructure and model development to user feedback and scaling. This full-stack approach has given us a deeper understanding of both the technical challenges and the impact our solutions can have.

Building strong relationships with our users

Along the way, we’ve learned just how important building strong relationships with our users is. During reviews and demos, it’s been inspiring to see customers actively engage with our solutions, offering valuable feedback and showing genuine interest in how these tools will help them. This connection to the end user has helped us stay aligned with real-world needs and continuously improve our products.

What’s crucial moving forward is our commitment to continuous learning. As we grow, we aim to stay true to our T-shaped approach: becoming experts in specific areas while maintaining a broad understanding of the entire stack. This balance will allow us to remain adaptable and innovative, while always keeping the bigger picture in mind.

A giant leap forward

Looking ahead, we’re energized by the possibilities that lie before us—new use cases, innovative features, and more ways to make a meaningful impact. What started as a single step into GenAI has grown into a giant leap forward, transforming the bank’s capabilities one GenAI building block at a time. It might have been one small step for our team, but we believe it’s a giant leap towards a future where AI-powered innovation is at the heart of Rabobank.

It’s been an incredible journey—from zero to hero—and we can’t wait to see what the next chapter brings. Here’s to creating more ground-breaking solutions, one step (and one leap) at a time.