Over the past few months, I have written about the hidden costs of building AI in-house, about what production-ready industrial AI demands, and about why the path from decision to operational value does not have to take 18 months. Those pieces were built on data, on research, and on patterns I observe across the many organizations I speak with.
But research and patterns only go so far. At some point, the most useful thing is to hear from people who have done it - who made the decision, went through the implementation, and are now living with the results.
Stefan van Bussel and Jeroen Wijnen at Berkvens Doorsystems are those people. Berkvens is a family-run door manufacturer based in the Netherlands, operating across three production facilities, and they were among the earliest adopters of Ultimo's digital workers. Their experience shaped how we think about what good looks like in practice.
The reason that debate matters at Berkvens is not abstract. Before working with Ultimo, they faced the same question every industrial organization faces right now: invest the time and resources to build AI capability in-house, or work with a platform that already carries the operational context, the domain knowledge, and the integration infrastructure. Their answer, and what they found on the other side of it, is what makes their story worth reading carefully.
What follows is their account, in their own words. I think it is the most honest description of industrial AI in production that I have come across - including the parts that are less often said, about data discipline, about the limits of what AI can substitute for, and about where the real value turned out to live.
What Happens When AI Joins the Maintenance Team
Every morning in a manufacturing facility, before a single machine is switched on, a maintenance team lead is already working. Reviewing overnight breakdowns, scanning shift logs, checking notifications, building a picture of what the machine park looks like before the day begins. At Berkvens Doorsystems, that process used to take between 30 minutes and an hour. Every day. Per team lead.
That time is now recovered by AI.
A manufacturer with nearly a century of tradition
Berkvens Doorsystems is a family-run manufacturer headquartered in Someren, Netherlands, and part of Xidoor. With three production facilities and five brands, we produce interior doors, frames, and sliding door systems for the housing, healthcare, hospitality, and education markets across Western Europe. Manufacturing at this scale demands disciplined asset management: diverse equipment across multiple production environments, high standards for uptime, and a technical team responsible for keeping it all running.
Like many manufacturers, we made a foundational investment in an enterprise asset management platform to bring structure to maintenance planning, work order management, and equipment tracking. That investment gave our technical department visibility it had previously lacked. But visibility only goes so far when the volume of data - notifications, shift logs, work histories, equipment records - keeps growing. At some point, reviewing all of it manually becomes the constraint. That is the problem AI is now solving.
What changed, and how quickly
We adopted Ultimo's digital workers as one of the earliest users. The starting point was practical: AI that analyzes the data the team was already generating, delivered through the EAM system already in daily use. No new platform. No disruptive implementation. Intelligence added to the workflow already in place.
The morning review process was transformed almost immediately. Where team leads had previously spent up to an hour manually cross-referencing breakdowns, notifications, and logs to build a status picture, the digital worker now assembles and summarizes that information automatically. The output matches the team's own manual analysis more than 95 percent of the time, meaning the time recovered is genuine, not traded against accuracy.
Across a team of technical leads, that amounts to between 125 and 250 hours of recovered productivity per person per year, with a direct financial value of roughly 12,500 to 25,000 euros annually per team lead. The math is not complicated, and it does not require a long build to get there.
Insights that were not visible before
Efficiency gains are the most measurable outcome, but not the most strategically significant one. What has mattered more is what the digital worker surfaces that manual review would have missed entirely.
By combining data across work orders, logbooks, and maintenance histories, the system identifies patterns that are difficult to detect when each data stream is reviewed in isolation. This revealed that certain equipment issues were recurring structurally rather than randomly - a finding that changed how the team approached maintenance planning and where it directed improvement efforts.
As Jeroen puts it: "By intelligently combining data from different sources, new insights emerge, enabling teams to set better priorities, identify structural issues, and carry out more targeted maintenance and improvements."
This is the practical meaning of intelligent asset management. Not AI as a reporting layer, but AI that changes which decisions get made, and when.
One lesson every manufacturer should hear
The most consistent message from our experience is that the value of AI is directly proportional to the quality of the data it works with. Disciplined, consistent registration of notifications, work orders, and operational data is a prerequisite - not something AI can substitute for. Poor data hygiene does not get corrected by AI. It gets amplified.
For manufacturers evaluating where to start with digital workers in maintenance, this is the practical readiness test. The technology is capable. The question is whether the data foundation is there to support it. For us, years of structured EAM usage meant it was.
What comes next
The area generating most interest for our team is troubleshooting support: a digital worker that not only identifies what has gone wrong but actively assists the diagnostic process, analyzing similar past failures and suggesting possible solutions in real time. For a manufacturer operating across multiple facilities, that kind of embedded guidance - turning accumulated failure history into decision support for any technician, regardless of experience level - addresses exactly the workforce challenge the industry is facing.
The conclusion we have reached is one more manufacturers are arriving at. AI does not replace the judgment and skill that experienced maintenance professionals bring. It removes the administrative burden that keeps them from applying it, and it finds the signal in operational data that no manual process reliably could.
The starting point is closer than most organizations expect. And the returns begin earlier than most anticipate.
What strikes me most about the Berkvens experience is not the time saved, impressive as that number is. It is how quickly the result arrived, and what was already in place to make that possible. Years of structured EAM data. A platform with the operational context already embedded. A team that did not have to spend 18 months building before they could start learning. That is the alternative this series has been arguing for. Berkvens is what it looks like.
