Embedding AI in Daily Banking Work: From Quick Wins to Scalable Support

The AI industry is obsessed with benchmarks and model sizes. However, in an enterprise environment like Rabobank, value doesn't come from a smarter model. It comes from solid integrations into daily workflows. At Rabobank's Conversational AI & CRM domain, we're working on this daily with more than a dozen live AI assistants. And there's more on the horizon.

While flashy AI assistants performing complex creative work grab headlines, we took a different, low-key approach: targeting "boring" yet high-volume administrative friction. We focused on embedding LLMs directly into daily workflows for summarization and knowledge retrieval to assist the employees. By deploying tools that summarize client conversations, our customer advisors benefit from a summary of contact histories and support from a knowledge assistant integrated with internal knowledge bases and process documentation systems.

The proof is in the volume. In December alone, we delivered over a million summaries to more than 20k unique users, supporting over 300k interactions. This means that if we only made the conversations on the helpdesk a single minute more efficient, we would save every person calling us a minute, and our employees combined 6,000 hours of tedious work.

Our biggest lesson is that technical capability is secondary to context. Success depends less on the model itself and more on the translation of business needs, connecting pain points and process realities with technical possibilities. Success requires a deep and pragmatic understanding of the human side of banking and mapping AI capabilities to those constraints.

Through Deep Dives and Value Discovery sessions with the operations department, we identified where Generative AI truly serves the business. These explorations were extremely useful, clarifying how this emerging technology can add tangible business value. Observing how we serve and interact with our customers at Rabobank provided critical insights into designing assistants, specifically defining what they can and can't do effectively.

Our latest development is an AI assistant engineered to handle activities in our CRM system. By utilizing prompt chaining, we linked multiple AI components together to create a summary of the current dossier plus retrieve relevant knowledge for the employee to review. This assistant reduces the time the employee spends "studying" the dossier to understand what they need to do. The case summarizer not only accelerates current workflows, but also establishes the architectural groundwork for future AI assistants, opening new possibilities for developing and scaling our assistants.

While model innovation continues, the real challenge has shifted toward embedding AI reliably into existing systems, governance structures and daily processes. Our focus is now on scaling what works. Deepening integration, improving trust and ensuring AI consistently delivers business value where it matters most.