AI decision support on the shop floor: practical gains or another layer of complexity?

Posted on 19 Nov 2026 by Joe Bush

Few still doubt the potential of AI within an industrial setting. Once deployed on the factory floor the technology has the capability to open many doors, significantly increase efficiency and visibility, and ease the burden on stretched workforces.

However, it is not merely a case of throwing technology at a myriad of manufacturing problems and seeing what sticks.

Careful consideration of the problem a manufacturer is trying to solve must be given prior to any deployment. Manufacturing is already grappling with a number of challenges, so the last thing that a business needs is a technology rollout that adds a further level of complexity to operations.

Here, The Manufacturer speaks to Harmjan Derksen, Chief Product Officer (CPO) at Ultimo, about how AI is adding value, where it’s struggling to deliver and the factors that manufacturers need to consider to get the best out of the technology.

Having hosted a number of roundtable discussions at the recent Manufacturing Leaders’ Summit, Harmjan certainly believes there is an element of untapped potential when it comes to AI. Yes, benefits are being seen across log books, Gemba boards, data analytics and the rapid creation of reports.

“However, technicians and operators are not really applying AI as I would have expected. So, there is a lot of room for improvement,” he said.

“During this event I’ve seen that AI is being used as a synonym for large language models and generative AI. But AI is of course much broader than that.

“The roundtable participants were also not sure about whether productivity gains can really be achieved without adding further complexity. It’s a point that software vendors and IT departments need to pay attention to. So, we’re working closely with the people on the shop floor to shadow what they are doing and improve their daily activities.”

AI faith and decision-making

Trust in AI is another challenge that was discussed at the event. What needs to happen before manufacturing teams feel comfortable acting on AI-generated recommendations, particularly when they could affect product quality and safety?

The answer comes down to when and how much we can trust AI. For example, if an AI system provides the correct recommendation 95% of the time but is wrong 5% of the time, where do we draw the line? What level of accuracy is good enough? We of course allow for human error – should the same leeway be granted to technology?

If AI consistently provides reliable recommendations, but humans are still expected to oversee every decision, there is also a risk of human oversight fatigue.

“People may become so accustomed to AI getting things right that they eventually stop questioning its recommendations. If an incorrect recommendation slips through because someone failed to spot the error, it could have serious consequences,” added Harmjan.

The key is to ensure that people remain actively involved in the decision-making process, rather than blindly relying on AI. Equally important is having a continuous feedback loop between people and the AI system. Users need to be able to indicate whether a recommendation was correct or incorrect, allowing the system to learn from its mistakes and improve over time.

While not there yet, Harmjan also envisages a day where, depending on compliance and regulations, AI is trusted to make fully autonomous maintenance-based decisions.

People may become so accustomed to AI getting things right that they eventually stop questioning its recommendations. If an incorrect recommendation slips through because someone failed to spot the error, it could have serious consequences.
Harmjan Derksen, Chief Product Officer (CPO), Ultimo

“Looking back 20 years ago, the same questions were being asked about computers and automation, so trust will improve over time.

“But it really depends on where you are on your journey and what decisions you’re taking. Is it high or low risk? What is allowed and what’s not? Companies in the EU will also have to comply with the EU Data Act.”

The importance of data quality as the foundation for AI also has a key role to play in its effectiveness. We’ve all heard the phrase ‘garbage in, garbage out’ and it’s true that without the right level of data quality, AI will only help an organisation make the wrong decisions faster.

However, it’s here where Harmjan believes there lies a real opportunity for AI to be applied to enrich and improve the quality of the data available.

He added: “You can use AI to collect more data to improve user experience in applications, for example, which lowers barriers and will help end users, operators and technicians to collect more data from which the organisation can then derive meaningful conclusions.”


AI decision support on the shop floor: practical gains or another layer of complexity?
The importance of data quality as the foundation for AI also has a key role to play in its effectiveness. We’ve all heard the phrase ‘garbage in, garbage out’ and it’s true that without the right level of data quality, AI will only help an organisation make the wrong decisions faster.

What will good look like?

Of course, another key consideration for any AI deployment is ROI. Deploying any digital technology across a manufacturing setting is no small undertaking, both in terms of financial outlay, and the time and effort that’s required during integration.

So, looking at AI through the lens of maintenance, can AI ROI purely come down to reduction in downtime or is the subject more nuanced than that?

Harmjan certainly thinks so. Both the keynotes on stage and the interactive polls that ran as part of Manufacturing Leaders’ Summit revealed that downtime is not as high a priority as one might first think.

Don’t talk constantly about AI, but about the problem you want to solve. AI, after all, is just a tool.
Harmjan Derksen, Chief Product Officer (CPO), Ultimo

“If you have an overcapacity, then downtime is not relevant,” Harmjan added. “AI ROI is also about cost effectiveness and your people; to provide them with new technology which can make their daily lives at the company better, which in turn improves retention. So, not all ROI can be expressed in hard British pounds.”

Key to any AI success is also how any initial project can be duplicated and scaled across the organisation. We often speak about ‘pilot purgatory’, where digital projects get marooned in their pilot phase and – regardless of their early promise and success – fail to scale up and provide similar benefits for the wider organisation.

In offering some best practice to avoid this pitfall, Harmjan advised manufacturers to not only start small – which is pretty understandable – but to start specific. Select clearly identified use cases and prove the benefits before attempting to scale.

“Don’t drop a bomb with AI because that won’t work,” he added. “You need to embed AI and make it part of your processes. It then just becomes part of the digital transformation journey.”

When asked if he could remove one barrier that is preventing AI from delivering greater value on the factory floor, Harmjan added: “Involve the people and select specific use cases. Select the problems which are there on the shop floor. Ask what the business is facing and what are the problems you want to solve?

“Then you can distinguish between the low hanging fruit to get results fast. So, don’t talk constantly about AI, but about the problem you want to solve. AI, after all, is just a tool.”

For more articles like this, visit our Data & AI channel