AI in industry: putting it into practice is proving difficult

Only 5 per cent of AI projects in industrial companies are currently of real value. This needs to improve. The 2026 Hannover Messe provided an overview of developments in artificial intelligence in Germany.Artificial intelligence (AI) offers businesses great opportunities for innovation, but for the time being, the results have been disappointingly meagre. That is the impression that lingers after speaking to experts at the Hannover Messe, one of Europe’s largest industrial trade fairs.Norbert Jung, head of industrial AI (artificial intelligence) at technology firm Bosch, recently highlighted an MIT study from mid-2025 in the engineering weekly *VDI nachrichten *. This research, carried out as part of the Nanda project (Networked AI agents in decentralised architecture), states that 95 per cent of AI projects in companies have no measurable added value.Jan Schaumburg, director of solutions and delivery at Insight Technology Solutions, is equally cautious. A survey conducted by his IT consultancy firm among 600 European technology managers shows that 80 per cent of their AI projects get stuck in the pilot phase. This is mainly because the projects are set up on too large a scale, whilst there is a shortage of skilled staff. Such staff must be able to navigate the existing IT infrastructure and also possess knowledge of modern AI technology. ‘That combination is rare,’ says Schaumburg.A solid foundation‘AI in industry’ was the theme of this year’s leading Hannover Messe trade fair last April. This was no coincidence. Germany sees itself as having excellent opportunities in this very field, thanks to its extensive and solid industrial base. AI can act as a catalyst for higher productivity, estimated by McKinsey at over 1 per cent annually. In its ‘High-Tech Agenda’, the German government has set out its intention that, by 2030, 10 per cent of gross domestic product should be AI-based. This will require billions in investment and appropriate training programmes.A solid foundation has already been laid with the ‘Industry 4.0’ project, launched over ten years ago, which placed a strong emphasis on automation. The idea is that AI, known in German as Künstliche Intelligenz (KI), can give German industry a competitive edge over companies from other countries, particularly China.There have certainly been successes to report. Robot manufacturer Kuka managed to reduce programming time by 80 per cent in a project using AI. The Volvo factory in Konz (Saarland) saved ten thousand staff hours with an AI project. Last year, Germany had 3,600 start-ups, some of which were specialising in AI software.According to board member Cedric Neike, technology group Siemens now supplies around 150 AI products, comprising both hardware and software. A Siemens AI tool analyses weld points at the Audi factory in Ingolstadt and controls robots for post-processing. Another product is being used at the Audi factory in Neckarsulm, where it is involved in quality control during the dip-coating of car bodies. According to Siemens, other projects have achieved productivity gains of between 20 and 40 per cent.Germany has around 25 ‘AI factories’ – large-scale data centres designed to accelerate industrial applications. In early February, the largest AI factory to date was commissioned in Munich on the initiative of Deutsche Telekom, representing an investment of one billion euros. Ten thousand Nvidia processors provide the computing power here. The hyperscale HammerHAI in Stuttgart focuses on developing AI solutions for manufacturing, engineering and research.Opportunities and limitationsAI has advantages and disadvantages, and presents both opportunities and limitations. There are many opportunities for innovation, but there are just as many limitations and threats. For example, ChatGPT offers many benefits when used properly, but there is a risk of misuse. Opportunities lie primarily in technology and science. AI can cut costs, improve machines and enable new products.But then there are the limitations. ‘AI is expensive,’ says project manager Aria Etemad from Volkswagen’s innovation department. This involves investment in hardware and software, but also in staff training. If it is no longer employees but AI models that take over the management of business processes, then the former will make less use of their professional expertise. However, this expertise is needed to assess and improve AI models.The European Union’s ‘AI Act’, which comes into force in 2024 and is primarily intended to protect consumers, is also perceived by manufacturers as restrictive. The business community is calling for greater freedom to exchange data between companies and with research institutions. At the opening of the trade fair, Chancellor Friedrich Merz pledged to advocate in Brussels for this request to be granted. The EU has since agreed to this.Start smallWhen setting up an AI project, companies must first determine what economic objective they wish to achieve, such as reducing material and/or energy consumption. Next, as with any automation project, it is all about organisation. When Christoph Knöll, CEO of consultancy firm Neurawork, starts an AI project with a client, he first discusses with management where bottlenecks in business operations lie, why turnover and profit are not higher, and why the organisation is not more efficient. ‘It then quickly becomes clear where the problems lie and how AI should be deployed.’Next, staff need to be retrained. Every company must acquire this knowledge to ensure AI is used correctly, not least to avoid being entirely at the mercy of the providers of this technology.It is wise to start AI projects on a small scale, in modules that can be integrated as the organisation grows. If you start with large, complex prototypes, they often get stuck at that stage.Machine learningMany AI applications essentially involve comparing new data with a very large database of older information, such as road sign recognition. This, incidentally, hardly merits the label ‘AI’. More advanced forms of AI, such as machine learning, are based on neural networks. These are organised in a way that mirrors the human brain. They consist of multiple layers of a series of neurons (processors), in which neurons from one layer pass data on to the next layer, and where the extent of data transfer can be adjusted for each specific application.Neural networks offer great potential for advanced forms of automation, but there are also challenges. These relate, for example, to reliability, traceability and the quality of the generated results. One persistent problem in particular is the ‘hallucinations’ of neural networks – the presentation of completely incorrect or implausible information as fact.Acting independentlyIndustrial AI only works with standardised data. Work planning, production, maintenance and service currently each provide their own data based on standards from different suppliers. But in the digital factory, everything is interconnected, so this is no longer feasible. The work planning laptops must therefore speak the same language as the robots on the factory floor and the maintenance department’s equipment.What a prompt is to AI programmes such as ChatGPT, the AI agent is to industrial artificial intelligence. However, this agent can do more than just provide a prompt: as well as supplying information, it can also act independently. An AI agent carries out autonomous actions to control robots and other machines. But it can also order materials, agree delivery times and adjust production parameters.Digital twinHowever, this powerful digital tool also carries certain risks. The quality of its work depends heavily on the scope and complexity of the tasks to be carried out. The larger the range of tasks and the more AI agents work together, the greater the chance that such a system will start to ‘hallucinate’ or become unstable. In the case of tasks assigned to machines or robots, this can pose a risk.The digital factory functions more effectively with work planning carried out by a digital twin, a software replica of the physical production process. This enables manufacturers to virtually design, test and optimise production lines. This can lead to significant efficiency gains: less time spent on set-up and commissioning, fewer prototypes and shorter production times.Human-like robotsAt the trade fair in Hanover, humanoids stole the show – human-like robots with arms and legs, fitted with sensors and 3D cameras. They have the potential to carry out tasks in the digital factory. ‘But not just yet,’ says Sven Parusel, director of research at the German-Chinese company Agile Robots, in a press briefing ahead of the trade fair. Moving around is no problem, but manipulating components remains a challenge. They are technically complex and, at a price of around eighty thousand euros, far too expensive. Elon Musk expects to be able to offer his humanoid Optimus for twenty thousand euros within a few years.‘Humanoids do not yet perform productive tasks,’ says Parusel. ‘We see wonderful demonstrations of walking, dancing and sport. These are truly major strides forward in research and development. But what we still hardly ever see is manipulation. And that is precisely the difficult part.’Empty phrasesAI plays a key role in the further development of humanoids. Research into walking systems and hands with fingers has been ongoing for some time. However, using traditional methods, it was extremely difficult to manage that level of complexity. ‘A five-fingered hand has an enormous number of degrees of freedom,’ says Parusel. ‘AI makes that complexity manageable. But as long as we do not have a firm grasp of manipulation skills, productive use in industry remains a distant prospect.’At ‘KI Talks’ in Hanover, many speakers praised AI as an opportunity for German industry. They were already talking about a ‘new industrial revolution’. But none of it was truly convincing. Particularly from the suppliers’ side, what came across was mostly marketing jargon and empty phrases: AI is ‘the greatest gift ever to the European economy’; ‘we are building the European counterpart to ChatGPT’. AI offers many opportunities for innovation, but putting it into practice is proving difficult. There is still a lot of work to be done. Photos: Deutsche Messe

AI in industry: putting it into practice is proving difficult

Read this article?

Meer artikelen

Moedermelkrobot verlaagt werkdruk op IC voor pasgeborenen

Drie consortia starten innovatierace emissieloze noodhulpsleepboten

Fietsbel is algoritmen te slim af

Databeveiliging voor de toekomst

Comments

Laatste artikelen

Moedermelkrobot verlaagt werkdruk op IC voor pasgeborenen

Drie consortia starten innovatierace emissieloze noodhulpsleepboten

Fietsbel is algoritmen te slim af

Meest gelezen

Read this article?

Sign up for our newsletter

AI in industry: putting it into practice is proving difficult

Read this article?

Sign up for the newsletter

Meer artikelen

Moedermelkrobot verlaagt werkdruk op IC voor pasgeborenen

Drie consortia starten innovatierace emissieloze noodhulpsleepboten

Fietsbel is algoritmen te slim af

Databeveiliging voor de toekomst

Comments

Laatste artikelen

Moedermelkrobot verlaagt werkdruk op IC voor pasgeborenen

Drie consortia starten innovatierace emissieloze noodhulpsleepboten

Fietsbel is algoritmen te slim af

Meest gelezen

TU/e-renovatietool helpt burgers bij verduurzamingskeuzen

RIVM: ervaren thuisklussers nemen vaker risico’s

Snelle wielwissel

Read this article?

Sign up for our newsletter