Two recent studies shed light on this dichotomy: the joint study by VDMA Großanlagenbau and Strategy&, and the international analysis ‘The GenAI Divide’ by MIT. Both show high expectations – but also clear discrepancies between ambition and implementation. The VDMA study shows that 84% of the companies surveyed rate GenAI as important or very important for their future profitability. On average, they hope to see an increase in EBIT margin of 6.7 percentage points – mainly through cost reduction rather than increased sales. However, despite this assessment, the actual rollout remains modest: AI is more than just a co-pilot and ChatGPT. Only 33% of companies with GenAI activities have implemented a company-wide, systematic rollout; the majority remain in pilot projects or the idea collection phase.
These figures reveal a contradiction between strategic importance and investment practice. The technology is there, as is the knowledge – but scaling and integration often fail due to a lack of resources, unclear strategy or cultural barriers. But let's not kid ourselves: engineering processes in mechanical and plant engineering are highly rule-based and therefore predestined for AI use.
The ‘pilot trap’: when use cases are not allowed to grow
Many companies remain trapped in the so-called ‘pilot trap’: individual departments test GenAI applications in project planning, engineering or sales, but the step towards company-wide integration is not taken. Yet use cases show great potential: from AI-generated technical documentation to automated quotation calculation, the approaches could show results within a few months.
According to the studies, the reasons for the hesitation are complex: there is a lack of specialised personnel, a clear target architecture and, in many cases, the digital infrastructure. Added to this are doubts about data quality and internal resistance to automation. The authors therefore recommend making GenAI a top priority and prioritising flagship projects with a clear return on investment.
The US MIT study comes to similar, but even more sharply drawn conclusions. It speaks of a ‘GenA divide’: while almost all companies are experimenting with generative AI, only about 5% of integrated GenAI pilots achieve measurable economic added value. The reason? Lack of context adaptation, insufficient learning ability of the tools and too strong a focus on generic large language models (LLMs) that cannot be integrated into operational processes. It is interesting to note that even companies with a high level of pilot activity are experiencing enormous scaling problems. Large companies in particular are finding it more difficult than medium-sized companies, whose more agile structure apparently enables faster implementation.
