AI projects don't fail because of the technology.

Most obstacles have nothing to do with technology. 66% of industrial companies don't want to hand their production data to external providers — that's a legitimate concern, not a sign of being behind the curve (BMWK 2025). Add poor or incomplete data (76% struggle, Bitkom 2025), no clear picture of what AI actually delivers (44% in mechanical engineering, VDMA 2025) and simply too little capacity for a year-long programme (51%, Bitkom 2025). The real problem isn't the technology — it's starting at the wrong point.

Five areas have proven themselves in German mid-sized manufacturing: automated visual quality inspection, early detection of machine failures, AI-assisted search through technical documents, analysis of production data, and automating office work. Just as important as the right choice: a pilot that isn't connected to your existing systems stays a toy — and makes it harder to win support for the next project. An honest note: without historical failure data you can usually only monitor the state of your machines, not predict when something will fail. That's still valuable, but a different starting point (Fraunhofer IPA). What matters is which approach fits your data, your equipment and your available resources — and that's exactly what this tool calculates.

Behind the result is a simple formula based on four questions: how many people or machines are affected? What's the economic value? How solid are the prerequisites — especially your data? And how much effort is realistic? From these four values we calculate one score. The higher, the more sensible the starting point. The result also shows whether a given approach would need dedicated hardware — a basis to discuss internally, not a conclusion after the fact.