The enthusiasm is palpable. Nearly two-thirds of the companies questioned in a survey conducted by the business federation Fedil, together with Luxinnovation and the Luxembourg Digital Innovation Hub, are in the most advanced stages of AI maturity. That means they have projects in development, in production or in structured experimentation phases. This maturity is particularly evident in sectors driven by technological innovation, such as ICT. On the other hand, the manufacturing industry--although interested by the new technology--is lagging behind. This is largely due to the practical challenges involved in collecting data in physical environments that are often not very digitised.
GenAI, for its part, is seeing rapid adoption: 70% of participating companies have already implemented use cases or are planning to do so. Tools such as ChatGPT and Microsoft Copilot are being used on a massive scale, with 87% of employees surveyed claiming to use them in a professional context. This widespread use of public tools is true even in companies that have formalised a governance policy: more than half of them nevertheless authorise the use of these solutions, often without effective control.
But challenges remain
As far as obstacles are concerned, the study highlights clearly identified roadblocks. Collecting reliable data remains a major obstacle, particularly in the transport, construction and industry sectors, where the physical environment and old equipment add to the difficulties. Amongst the least advanced companies, only 24% have operational data collection capabilities; 62% have not yet implemented basic data collection functions.
Lack of in-house expertise tops the list of challenges cited. The difficulty of identifying relevant use cases reinforces this shortcoming. And despite the existence of public initiatives such as the Digital Learning Hub, 61% of respondents are unaware of its training offerings. Similarly, academic institutions and consultants are much less visible than Luxinnovation or AI solution providers.
A major challenge lies in the concealment of government support mechanisms for AI; the survey results indicate that these mechanisms do not effectively reach their intended beneficiaries. There is a critical communication gap that requires urgent attention. The level of awareness of available support mechanisms is low, with only 20% of respondents being aware of them. Even fewer are able to identify a specific mechanism.
On the issue of digital sovereignty, only 38% of respondents are planning local development or hosting of their GenAI solutions. The majority rely on external tools, which brings the risk of creating technological dependency and compromising control of sensitive data.
Finally, although 56% of companies say they have a data and AI governance policy, it often remains disconnected from actual practices. Three out of four companies still have not anticipated the requirements of the European AI Act, despite the fact that it is already being implemented.
This article was originally published in French.
