Insights from All About Automation Zurich: Why AI Projects Fail Without Structured Data
At this year's All About Automation exhibition in Zurich, Michael Meisel shared insights on "Why AI Projects Are Bound to Fail – Structured Data as the Foundation."
During the discussions afterwards, three production managers agreed on the same point: much of the current AI hype doesn't hold up under real shop floor conditions. Ask an LLM the same question five times, and you get five different answers.
Where Machine Learning and AI Actually Deliver Results
The production managers we spoke with agreed: successful AI use cases aren't built on unstructured text queries. They're built on structured, high-quality machine data. Three examples show what this looks like in practice:
Pattern recognition in high-volume data. In high-vertical manufacturing, manually reviewing rows of numerical data isn't feasible. Machine learning algorithms scan these data streams automatically, identifying correlations and deviations that would otherwise go unnoticed.
Reducing scrap in rework. Knorr-Bremse useds AI-powered error analysis to guide operators directly to the root cause of production issues, reducing the rework error rate by up to 30%. Link to case study
Reducing MTTR with structured data linking. Working with an international electronics manufacturer, we aim to connect logbooks, maintenance logs, error messages, and system manuals to unlock previously unstructured data. To help operators with precise troubleshooting recommendations of a malfunction, significantly reducing Mean Time to Repair.
Three Questions Worth Asking About Your Production Data
Conversations at our booth made one thing clear: most plants already have data. The challenge is what condition that data is in.
-
Data quality. Is your data actually structured well enough to support automated analysis, or is it, in practice, unusable "data garbage"?
-
Procurement specifications. A generic "data transfer interface" requirement in your technical specs isn't enough. Structured data needs to be explicitly defined, demanded from the machine builder, and validated at final acceptance.
-
Manual aggregation. Many companies still compile production data by hand. This means no real-time visibility, delayed decisions, and more opportunities for error.
Next Steps
STIWA Shopfloor Software was built by manufacturing experts based on direct, hands-on experience with production data — from tapping into the control level to turning that data into measurable productivity gains.
If you'd like to evaluate your current data quality, book a short call with the team.
