49th FVI Expert Breakfast

Topic: "What remains when the machine calculates?" – Why AI does not replace human experiential knowledge but even increases its importance.

The session deliberately did not focus on individual AI tools, but on the question of how companies can work with AI without losing their own decision-making ability. It became particularly clear: Data and AI can recognize patterns – but the crucial context often still resides in the minds of experienced employees.

• The machine knows the data – the human knows the context: A particularly illustrative example was a supposedly trouble-free running drive. Sensor data, temperature, and history looked inconspicuous, which is why an AI might recommend extending the maintenance interval. However, the experienced maintenance technician disagrees because he knows the actual load profile, frequent start-up processes, and additional thermal stresses. His achievement is not in calculating better, but in correctly classifying the validity range of the data. This is where human judgment lies.
• Securing knowledge is not enough – companies must secure judgment: When experienced employees leave, documentation, databases, or wikis are often created today. This saves rules and facts, but often not the crucial exceptions: Why was a conscious decision made to act differently in a particular situation? What warning signs lead to contradicting a system recommendation? The central insight of the episode is therefore: Companies must not only preserve knowledge but also capture justified deviations, boundary conditions, and experiential judgments.
• AI should prepare decisions, not replace responsibility: The most succinct sentence of the session was: "The human decides, the machine calculates." AI can analyze, establish connections, provide variants, or create solution proposals. However, the responsibility for the decision remains with the human. It becomes particularly critical when AI-generated results are passed on unchecked through several hierarchy levels, leading to a diffusion of responsibility: In the end, the result appears professional, even though the original question may have been wrong.
• Efficiency should not simply lead to even more work: Another important point became clear in the discussion: If AI saves time, the gained efficiency should not automatically mean that employees simply receive even more tasks. Instead, free time can be used to improve processes, document knowledge, further develop systems, and try out new solutions. At the same time, AI creates a new problem: If every employee can generate large amounts of reports, analyses, and proposals in a short time, the bottleneck shifts from creation to evaluation and responsibility. Quality and conciseness thus become more important than pure output quantity.

Classification: From stored knowledge to Operational Intelligence

This episode fits almost ideally with our strategy at Hahn PRO: ADAM should not replace the experienced technician, but make his knowledge, context, and judgment usable and permanently available in everyday operations.

• Experiential knowledge resides in the head rather than in the system: The episode clearly shows the core problem: Databases often know the normal case, but not the exception. However, it is precisely this exception that often determines in maintenance whether a system continues to run or comes to an unplanned stop. With ADAM, we turn this implicit knowledge into a digital corporate asset. Through voice-based feedback, technicians can not only document what they have done but also why. This information flows into the Digital Life Record of a system. Thus, a form of Operational Intelligence is created step by step, connecting machine data, documentation, and human experience. ADAM thus becomes the digital vault for the maintenance experience treasure.
• Too much data, but too little usable context: In many brownfield environments, information is distributed across ERP systems, documents, tickets, folders, and the minds of individual employees. The problem is therefore not necessarily a lack of data, but its findability and classification. Our approach is ADAM-First and Brownfield-Native: We do not wait for all master data to be perfectly cleaned and every legacy system unified. ADAM connects existing information and makes it directly usable through intelligent search and AI assistants. The technician does not need to know in which system a piece of information is located; he needs the right information about the system and the specific malfunction at the right moment.
• AI must not push humans out of the decision-making process: The discussion about liability and responsibility hits the core of our product philosophy. A black box that makes decisions independently would be the wrong approach in many operational situations. We therefore understand ADAM as augmentation technology: as a digital colleague and not as a digital superior. The solution assistant can gather historical cases, documents, and known solutions. However, the human evaluates this information based on the real situation on-site. At the same time, the feedback assistant ensures that this new experience is structured back into the Digital Life Record. Thus, a continuous cycle of searching, deciding, acting, and learning is created.

Conclusion: The better machines can calculate, the more important becomes the human ability to classify, question, and take responsibility for results, and for this, systems like ADAM are needed, which not only collect data but transform human experiential knowledge into Operational Intelligence.