A cross the world, manufacturers face mounting pressure to boost productivity, innovate faster and stay competitive in an increasingly volatile environment. Agentic AI is helping individual users to do more – but according to the Massachusetts Institute of Technology, in 95 per cent of cases it still fails to deliver economic impact for the business.
For Norbert Jung, CEO of Bosch Connected Industry, and Rupert Stuetzle, general manager of EMEA manufacturing and mobility at Microsoft, it’s the quality of the AI program – not the quantity of agents deployed – that counts. A combination of technology, knowledge and context is the only way to deliver meaningful results.
What changes are you seeing in the way manufacturers approach AI?
RS: We’re seeing a sense of urgency in the companies we talk to. Two or three years ago, AI was mainly a topic for experimentation, but now this is changing. Companies are facing a lot of challenges – cost pressure, competitive pressure, regulation, disruptions in the supply chain – and they are looking for solutions. They’re no longer looking to experiment with AI; they want to know it will deliver impact at scale. The question is, how can we achieve real business impact?
Every larger company has AI pilot initiatives. The challenge for manufacturers is how to get those pilots to production and implement them at scale in a real-world factory.
Why are so many AI deployments failing to deliver business value?
NJ: There are several reasons, from slow time to value to limited scalability. We often see a pilot-to-production gap, where companies think the pilot will do the job, but when they start to roll it out, things become more difficult. Pilots don’t immediately show the costs of operating, maintaining, integrating and scaling those agents, and the change management needed to drive user adoption.
What is the ideal foundation to enable AI at scale?
NJ: Strong partnerships are essential. At Bosch, we compared self-build deployment models with those involving partners who have industrial domain know-how, AI competence and data. We found that firms working with an ecosystem of partners can deliver value 74 per cent faster than those that go it alone. They will also typically yield 25 per cent more value, with 30 per cent less cost.
What’s the best way to measure the success of agentic AI, and how can manufacturers achieve it?
NJ: We see a lot of companies that pride themselves on the number of agents they have deployed – but success with AI is not about resource consumption; it’s about economic outcome. To evaluate that, you need a scalable system that produces trustworthy operational intelligence and measurable economic impact.
Context is the key to producing that intelligence. Take root cause analysis for example. Quality engineers rely on many different source systems – from engineering drawings to production and end of line tests – to conduct these assessments, and each system describes the product in a different way. On its own, a large language model (LLM) doesn’t know how to deal with that. In fact, we asked quality engineers if they were happy with standard agentic AI systems’ responses to 34 typical daily job questions. Only 20 per cent of cases met with approval.
However, if you deploy an out-of-the-box LLM on semantic data, which brings all those source systems into a meaningful context, the quality of response increases to 60 per cent. And if you apply industrial manufacturing domain know-how, it increases even more.
Manufacturing Co-Intelligence brings together humans, machines, AI and data to unleash the next wave in operational intelligence (Photo: Robert Bosch GmbH)
How does Manufacturing Co-Intelligence bring together the context and know-how manufacturers need?
NJ: Manufacturing Co-Intelligence is about unleashing the next wave in operational excellence with industry expertise from Bosch. We bring together humans, machines, AI and data to do that.
One example is our shop floor agent crew, which is deployed in Bosch and in external customers to help with a critical problem in manufacturing – the scarcity of skilled operators on the shop floor. Our agent provides responses of equal quality to a senior expert. The system is rolling out to operations around the world, and in 60 per cent of cases, users say their use of the agent has directly reduced machine downtime. This translates to almost a €1 million ($1.14 million) benefit for a typical plant of around 2,000 workers.
Bosch’s domain know-how is necessary to bring that economic impact and trustworthiness to agent systems, but so is the technology platform behind it. We could not do this alone; no company can.
RS: Manufacturing Co-Intelligence is a combination of technology and manufacturing know-how. The challenges in manufacturing are all about processes and dependencies. They’re about legacy systems, and there are many things that generic AI doesn’t really understand that well. Bosch brings decades of experience in manufacturing. It helps to provide the semantic context that drives true solutions – not just intelligent answers based on inference, which a standard AI system can do, but real operational decisions.
What makes your own partnerships so successful in unlocking the value of agentic AI for manufacturers?
RS: It takes multiple elements to solve the industrial AI challenge. At Microsoft we provide the technology foundation – the cloud and AI platform, the security, the global scale. We also talk to our customers to understand their business challenges, and we think about how to solve them by combining our competencies with the expertise of partners who are deeply ingrained in the industrial space.
That’s where Bosch is uniquely positioned, because as well as being a software provider, it operates a global manufacturing system. It’s clear that Bosch is using these solutions itself, and that makes a big difference for customers.
NJ: We work in an ecosystem of companies. Alongside Microsoft we have Bain & Company, which is helping to shape the future vision of manufacturing. In addition, Capgemini and Deloitte help us drive IT programmes and transformation within our customers, including change management, which is vital to drive user adoption. Databricks, as a native Azure service, is the prime system to build our semantic data layer on.
Each partner brings something unique and valuable that strengthens our shared mission: to drive operational excellence for our customers.
Read more like this, including news, interviews, case studies and insight from industry experts, in the Autumn 2026 issue of Technology Record. To get content like this on a regular basis, subscribe to the print edition or free digital edition of the quarterly Technology Record magazine.