"Effective AI happens when everyone involved sits down together."

Oliver Jung, Head of Corporate Development, and Data & Analytics expert Tim Graf report on how they are establishing viable guidelines and a common understanding for the implementation of AI projects at BARMER. Together, they explain why small steps, open dialogue, and clear expectation management are crucial to unlocking the effectiveness of AI.

Many companies are currently facing the challenge of integrating generative and agentic AI into their processes in a meaningful way. Where do you see the biggest barriers?

Tim Graf: One major hurdle is the regulatory environment—especially in the public sector. When it comes to cloud-based solutions in particular, we have to examine very carefully how we can implement AI in a way that complies with data protection regulations. Then there is the question of responsibility: Who is responsible for governance – especially with regard to the EU AI Act or geopolitical dependencies? This must be clearly defined in advance. From a technical perspective, too, fundamental decisions have to be made: Which services do we purchase, and what do we develop ourselves? The architecture must be set up quickly but sustainably. In order to make these decisions, it is first necessary to establish a common understanding of what AI should and can achieve. Everyone wants to do "something with AI," sometimes without knowing the specific use case. This requires technically strong teams that are able to apply AI to real-world problems. Only then does it become a strategic tool that can be used in a targeted manner.

Oliver Jung: I completely agree – that's precisely why a kind of phase 0 is needed before the project starts, in which the basic framework conditions are defined. What we also learned last year is that these parameters never remain static. When working with AI, we must be prepared to question assumptions again and again. This insight is important, especially with regard to expectation management – including at the management level. Management commitment is often the biggest hurdle. Many people still think in terms of classic project logic, but AI requires a more agile approach. It is better to move forward in small, tangible steps than to wait for the one perfect solution. And risks cannot be addressed on the whiteboard, but by getting them tested as quickly as possible and putting them on the table. This approach must be clearly communicated and agreed upon in advance.

How did you proceed to ensure that this transparent communication and integration was successful?

Oliver Jung: A key point was to build trust. To do this, we first had to understand who had doubts and why. Transparency begins with listening: What are the concerns? What assumptions are being made? We deliberately conducted the dialogue in small, structured formats – workshop by workshop, always focusing on specific issues. This helped to make the complexity tangible. It was also important to involve management from the outset – especially those who are responsible for implementation and have high expectations of the results. Taking their perspective into account at an early stage significantly increased acceptance. In many discussions, we established a common understanding that it is perfectly acceptable to start with assumptions and to decide how to deal with them if they later prove to be untenable.

The full interview with Oliver Jung and Tim Graf can be found in our white paper "Road to Agentic AI. The fascination of automation."