23rd FVI Expert Breakfast

Topic: "Aimlessly through the AI Galaxy?" – A guide for the structured introduction of AI in maintenance.

Guests included Prof. Lennart Brumby and his students Kai and Baris (DHBW Mannheim), who presented a guide to prevent AI projects from failing.

• The 80% Failure Rate: 80% of all AI projects fail. Why? Not because of technology, but due to the wrong choice of problem ("We have to do AI now, no matter what for") and lack of data quality.
• The "Stop-or-Go" Approach: The presented guide recommends strict termination criteria ("Stop or Go"). If the profitability is not clear in phase 1: Stop. If data is missing in phase 2: Stop. Many companies continue even though the basis is missing, burning money.
• Low Hanging Fruits First: Instead of starting with the most complex predictive maintenance project (which requires 10,000 sensors), one should begin with simple use cases. Example: Knowledge management. Making documents searchable (RAG) is easy, brings immediate benefits, and doesn't require a perfect data history.
• The Stakeholder Trap: Tina (IT service provider) emphasized: You have to involve the users (workers) from day 1. If the GUI (user interface) is poor, the best algorithm in the backend is useless. Acceptance determines success.
• Data Quality is a Process: Fabian (software manufacturer) warned: "AI doesn't fix bad data." If values were entered incorrectly in the lab (typos), the AI learns nonsense ("Garbage in, Garbage out"). You need experts to validate the data before training.

Classification: Structured introduction instead of rampant growth

This episode confirms our approach of guided introduction ("Customer Success").

• ADAM's "Fast Lane": Our offer aligns with the experts' recommendation: Start with a "Low Hanging Fruit" (knowledge transfer/RAG) that quickly brings ROI, instead of a huge cloud project. We don't sell a "Big Bang," but step-by-step success.
• Data Validation through Dialogue: We solve the problem of "bad data" through dialogue. If a value is implausible, ADAM asks: "Are you sure the temperature is 500 degrees? That would be very high." This creates data quality at the source (with humans).

Conclusion: AI is not a project, but a process. Those who proceed in a structured manner and involve employees belong to the 20% who are successful.