Few topics dominate the industrial discourse today as strongly as Artificial Intelligence (AI). Buzzwords such as Predictive Maintenance, Digital Twin, or Smart Factory appear in almost every strategy presentation. Yet, the gap between ambition and implementation is often wide. Many plants already have vast amounts of data, but only a few use it consistently to derive real insights. The reasons are varied: poor data quality, unclear objectives, or simply the fear of overwhelming complexity.
Albert Schiller, Managing Director of XPNX, shares his view on how AI can create tangible value in manufacturing – when it is properly understood and applied.
Mr. Schiller, how do you see the current state of AI in manufacturing?
AI is everywhere today – at least in presentations and strategy papers. In reality, however, it is still in its infancy in many places. Many companies have started collecting data, but that alone does not mean they have data competence. AI is not a miracle cure; it is a tool, and it is only as good as the data you feed into it.
In practice this means: first comes process understanding, then the application of algorithms. If a plant does not understand how its processes interconnect, even the best models will not help. Many organizations invest in technology before they define which specific questions they want answered – for example: Why are downtimes occurring? How are temperature, pressure, and scrap connected? The sequence is crucial: first the goal, then the method, then the tool.
Another common misconception is that AI is seen as a replacement for human expertise. In reality, it works best as a complement. An experienced production manager can intuitively recognize patterns, and AI can confirm or quantify those impressions with data. This creates a powerful interplay between experience and calculation. AI does not replace expertise – it amplifies it.
Where do you see the most meaningful applications?
Wherever patterns, repetition, and large datasets are involved. Classic examples are predictive maintenance and quality forecasting.
In predictive maintenance, the goal is to detect from sensor data such as temperature, vibration, or current draw when a machine is likely to fail. This sounds straightforward, but it requires clean data, historical records, and an understanding of what “normal” looks like. Only if you know the baseline can you identify anomalies early.
A second strong use case is quality forecasting. AI can detect which combination of process parameters leads to stable results – for example, which temperature ranges or filling levels reduce scrap. It can uncover patterns that are invisible to the human eye because they stretch across long timeframes or multiple variables.
Beyond that, there are many additional fields: energy optimization, production planning, recipe control, supply chain analysis. What matters most is always the right question: What do I want to know, and what can I influence? AI can show correlations, but it cannot take responsibility. Interpretation remains with the human expert.
Another important point is expectations. AI rarely delivers spectacular results overnight. Its real value lies in making hidden relationships visible. That creates understanding – and understanding is the foundation of every improvement.
How can companies avoid the “buzzword effect” around AI?
By not treating AI as a project, but as an integral part of daily operations. AI must not be an end in itself; it has to be embedded into a larger system – processes, responsibilities, and decisions. If it remains isolated, it quickly becomes a prestige object: expensive but ineffective.
A common mistake is that AI projects begin in innovation departments but never reach the production floor. Insights need to arrive where the data is generated. That means AI has to be understandable. If an algorithm delivers results that no one can explain, nobody will trust them. Transparency and simplicity are essential – explainable models, clear visualization, and the ability to reconcile outcomes with human experience.
Clear roles are also necessary. AI changes not only technology but also organizations. Who is responsible if a model makes the wrong recommendation? How are human decisions weighed against machine suggestions? These questions determine whether AI will be used sustainably or remain a one-off experiment.
In the end, it all comes down to culture. An organization that understands, maintains, and leverages its data will succeed even with simple models. An organization without data awareness will achieve little, even with the most advanced technology.
What prerequisites must be met for AI to work in manufacturing?
Conclusion
Artificial Intelligence unfolds its value not through its mere existence, but through its application. It becomes a success factor when it supports decisions rather than replaces them. That requires more than technology – namely a clear objective, clean data, and an organization ready to learn from insights.
Outlook
In the years ahead, AI will become a standard part of industry. But the difference between success and stagnation will continue to lie in execution: Not those with the largest datasets will win, but those who ask the right questions – and understand the answers.
