Between Regulation and Industrial Implementation: Schaeffler Focuses on AI Quality
MISSION KI is developing a voluntary quality standard for AI systems that fall below the high-risk threshold of the EU AI Act, to be completed by the end of 2025. A circle of experts from industry and research is accompanying this process with critical and constructive input.
Schaeffler is part of this expert group. The company applies AI in various domains—from production optimization and digital twins to generative AI platforms. In all these applications, quality assurance plays a crucial role, as inaccurate predictions, for example, can lead to production downtime.
We spoke with Dr. Thomas Klein and Dr. Dominik Riedelbauch from the Advanced Innovation division at Schaeffler, who have accompanied the development of the standard as members of the expert group, about industrial needs, practical challenges, and realistic expectations.
What motivated you to take part in the MISSION KI expert group?
At Schaeffler, we are working with AI in a wide variety of areas—from intelligent control and condition monitoring to AI assistants. Reliability and quality are essential. If, for example, predictive maintenance systems fail to detect faults in time or AI-based programming assistants produce faulty code suggestions, this can lead to operational issues. That’s why we find the exchange on AI quality requirements both relevant and stimulating.
What specific contribution did you bring to the expert group’s work?
Our perspective is primarily technical and method-oriented. We are familiar with a wide range of AI technologies across different contexts—which also means diverse concepts of quality. In the discussions, we focused on critically assessing the technical feasibility of the measurement instruments proposed for the various quality dimensions: Are the procedures clearly defined? Are implementations available? How high is the additional effort? Which testing methods are actually practical in industrial environments?
How important is AI quality for industry—today and in the coming years?
Clear quality requirements form a fundamental basis for the efficiency and reliability of AI-supported processes and products, both now and in the future. This applies equally to internal applications and to AI-based solutions offered to customers.
Where do you see concrete needs at Schaeffler with regard to AI quality? What do you expect from the new quality standard?
Today, there is already a wide range of methods available for addressing AI quality—for example bias detection, robustness testing, and explainability approaches.
In practice, however, applying these methods can be complex—for instance when selecting suitable test data, defining meaningful metrics, or interpreting results. Current research literature provides valuable insights but often lacks clear implementation guidelines.
The new standard can therefore serve as a practical point of orientation, offering hands-on tools that make it easier to define and verify quality requirements.
What advice would you give to SMEs that are just beginning to address AI quality?
We recommend choosing an existing AI quality framework and testing it against your own use cases: What relevance do the defined criteria and measures have for your specific applications, and how can they be implemented in practice? This process can be demanding, but it greatly enhances understanding of how AI quality can be applied in real-world contexts. Ultimately, what matters most is not which framework you choose, but how deeply you engage with the topic.
