AI in manufacturing: less office work, more making

The AI we build for manufacturing doesn't do the work at the machine, but the work around it: reading requests and drawings, preparing cost estimates, checking order confirmations, proposing schedules, writing quality reports and making the knowledge of experienced colleagues searchable. The work preparation engineer, estimator and planner decide. AI does the searching and typing.

Industry is still cautious about this. According to the AI monitor of Statistics Netherlands (CBS), 18 percent of industrial companies with ten or more people employed used one or more AI applications in 2024, against almost 23 percent of all companies of that size. That's not a lack of opportunities, but a lack of time: if you have to get orders out the door every day, there's no room to experiment.

At the same time, the pressure is high. At the end of the third quarter of 2025, the Dutch Employee Insurance Agency (UWV) counted 3.5 times as many vacancies for industrial occupations as people who had been on unemployment benefit for six months or less. You can't easily replace the experienced estimator who is retiring. You can capture their knowledge.

By capturing their knowledge where it already is: in past orders, cost estimates, drawings, deviation reports and in their heads. AI makes that searchable and usable for the colleague who takes over the work. The expert remains the source, the assistant becomes the memory.

In practice it works like this. We collect the documents that already exist and interview the experienced colleagues about how they decide: why this operation first, why this margin on this material, what to watch out for with this customer. Together with hundreds of past orders, that becomes the basis for an assistant that answers questions and prepares a cost estimate the way Henk would.

The nice thing is that Henk is often enthusiastic about it himself. He no longer has to answer the same question ten times a day, and his craftsmanship stays in the company after he leaves.

What you need

• Past orders and cost estimates, preferably going back a few years
• Drawings and bills of materials in a place where you can find them
• A few hours from the experienced colleagues, spread over a few weeks
• Someone who checks the assistant's answers in the first months

AI on the shop floor itself, such as cameras that inspect products or sensors that predict when a machine will break down, is a field of its own. It requires sensor data, hardware and specialists in that technology. There are good companies for that, and we're not one of them.

We focus on the work around production: everything to do with reading, calculating, planning, checking and writing. In most manufacturing companies there's more to gain there than you'd think, because it doesn't happen on the floor and so doesn't show up in the OEE figures.

And if it turns out that a standard package solves your problem, for example a planning module your ERP already has but nobody uses, we'll say so. That's faster and cheaper than building.