As AI copilots move into maintenance and operations workflows, manufacturers must separate measurable downtime reduction from digital noise. Here’s what early adopters are learning.
Ask a room of operations directors about artificial intelligence (AI) and you will hear two conversations happening at once. One is about genuine results: hours recovered, failures caught earlier, knowledge captured before it walks out of the door. The other is about fatigue, with another dashboard, another pilot, another layer between the technician and the job.
Both conversations are legitimate. The difference between them usually comes down to where AI has been applied, and whether anyone bothered to measure what changed.
The stakes are not abstract. Research by Splunk and Oxford Economics puts the cost of unplanned downtime to Global 2000 companies at $600 billion a year, a figure that has climbed 50 per cent in two years. ABB estimates the average hourly cost of unplanned downtime across industrial sectors at nearly $170,000. Against numbers like that, “wait and see” is itself an expensive strategy. But so is deploying AI that nobody trusts or uses.
Start where the pain is measurable
The AI deployments delivering value in manufacturing today share a common trait. They attack a specific, quantifiable problem rather than promising general transformation.
Take Berkvens Doorsystems, a family-run door manufacturer operating three production facilities. Before AI entered the picture, each maintenance team lead spent between 30 minutes and an hour every morning cross-referencing overnight breakdowns, shift logs and notifications just to build a picture of the machine park. That work is now assembled automatically by an AI digital worker embedded in their existing asset management system, and the team has verified that the output matches their own manual analysis more than 95 percent of the time. Across a team of technical leads, that translates to between 125 and 250 hours of recovered productivity per person, per year.
Notice what made this work. The problem was narrow and well understood. The baseline was known, so the gain could be verified. And critically, the AI was added to a workflow and a system the team already used every day. No new platform, no parallel process, no extra screen competing for attention. That last point matters more than most technology discussions acknowledge. AI that lives outside the daily workflow is complexity however clever it is.
The quieter value: knowledge that stays
Efficiency is the headline benefit, but it may not be the most important one. Sixty-three per cent of industrial organisations report struggling with an ageing workforce, according to our Maintenance Trend Report, and the deeper problem is not headcount. It is that decades of diagnostic intuition, such as knowing how a particular wrapping line behaves on a humid day, exists in people’s heads and nowhere else.
This is where food and beverage producers, who face some of the harshest downtime economics in manufacturing, are moving early. Steenland Chocolate, a manufacturer of customised chocolate coins exporting to more than 50 countries, deployed AI digital workers to automate maintenance planning and surface historical repair knowledge at the point of work.
“With AI functionalities in Ultimo, we improve the quality of our work: failure reports are more accurate, work orders are smarter, and maintenance knowledge is better secured,” says Kevin Brusselaars, manager of projects and technical services at Steenland. His caveat is just as instructive: “Data quality and human involvement remain essential.”
That caveat should be printed and pinned to the wall of every AI steering committee. At Berkvens, the team reached the same conclusion independently. That is that AI does not correct poor data hygiene, it amplifies it. Disciplined registration of notifications, work orders and shift logs is the prerequisite, not the afterthought.
Trust is earned in specifics, not asserted in slideware
The hardest question is not technical. It is whether a technician with 20 years of experience will act on a recommendation from software, and whether they should.
In my experience, trust on the shop floor is built on three things. First, transparency: people need to see why the system recommended what it did, and what data it drew on. Second, boundaries: routine, low-risk actions can be automated, but higher-risk decisions must escalate to a person, with every action audit-logged. Third, verification: the Berkvens team trusted their digital worker because they checked its output against their own analysis for long enough to know it held up. Trust was measured into existence.
Organisations that skip this stage and simply mandate adoption tend to end up with the worst of both worlds: the cost of AI and the workarounds of people who quietly ignore it.
Where to start
For manufacturers weighing this up, four questions cut through most of the noise. Is your maintenance data disciplined enough to be worth analysing? Can you name a specific decision or task the AI will improve, with a baseline to measure against? Will it live inside the systems your team already uses? And can your people see, question and override what it does?
If the answer to all four is yes, the returns tend to arrive earlier than expected. If not, the honest move is to fix the foundations first, which is unglamorous work, but it is where every successful deployment I have seen began.
The debate about AI on the shop floor is worth having out loud, with practitioners rather than vendors doing most of the talking. That is precisely the conversation we are convening at our upcoming roundtable, AI Decision Support on the Shop Floor: Practical Gains or Another Layer of Complexity?, where manufacturers will share where AI is delivering measurable value, and where it is not. Because the technology has stopped being the hard part. Judgement is.
Candi Robison, VP Enterprise Asset Management Strategy & Innovation, Ultimo
Candi Robison utilises more than two decades of experience in asset management and reliability to transform manufacturing, energy and logistics businesses in the Americas for Ultimo, the AI-augmented enterprise asset management company. She shapes the company’s long-range global EAM vision – working across marketing, sales, product, sales and customer success disciplines.
For more articles like this, visit our Industrial Data & AI channel.
