Smarter Operations Through Engineering

Manufacturing has long served as a proving ground for emerging technologies, from augmented reality headsets guiding engineers to machine vision systems for quality checks and robotic arms building other robots. This track record often positions the industry as an early adopter of innovation. Yet the gap between those flashy case studies and the typical factory floor remains vast. “You get some really exciting examples, but it’s not pervasive,” notes Chris Dungey, CTO at High Value Manufacturing Catapult. The complexity of manufacturing processes, with many moving parts and high safety stakes, has slowed widespread AI integration despite the clear opportunity.

Back-office operations face similar hurdles. In 2024, the Manufacturing Leadership Council found that 70 percent of manufacturers still relied on manual data entry. AI adoption in knowledge work has been sluggish because these roles often lack rules-based structures, making automation inherently harder. However, the need is no longer optional. “We’re having a rough economic time here in Europe right now, and what we’ve heard from companies is, ‘We want to do more, but with the same staff,'” says Alexander Müller, cofounder of Workist, which builds AI software for white-collar tasks. While full-scale transformation remains elusive, early movers offer valuable lessons.

Lesson 1: Keep Humans in the Loop

Michelin digitized much of its factory operations by 2020, but administrative and knowledge work stayed manual. Since then, the company has focused on augmenting workers with AI rather than replacing them. After early experiments in 2021, Michelin built its data foundations and now scales AI across more than 200 business cases. Ambica Rajagopal, group chief data and AI officer, intentionally avoids restricting which processes can use AI. “The key to adoption of AI, for me, is empathy,” she says. Employees can use an internal generative AI tool called Aurora, which supports knowledge agents trained on team-specific documents for finance, legal, and other departments. Workers can also create their own self-service agents, and some have developed unexpected use cases independently, like comparing information across documents in different formats.

At Harting, a global industrial manufacturer, team leader Marcel Nattke saw similar enthusiasm by involving employees early. He let teams trial multiple solutions and ensured the key user wasn’t a manager, empowering lower-level staff to become tech liaisons. Transparency was built into the chosen software, Workist’s order processing tool, which uses a traffic light system: green for full automation, yellow for human review, and red for failure. This builds trust, says Müller, giving non-specialists a glimpse into the AI’s decision-making. Nattke recalls the project succeeded when a colleague said, “I feel that the system is lifting the weight off my shoulders; I have time to breathe again.” On the factory floor, Dungey echoes this: “If the engineers, operators, and supervisors do not trust the system it will not scale. It will be a quiet rejection.” He stresses that buy-in must span the entire organization, from shop floor to top floor, with upskilling to avoid what he calls the “graveyard of tools.”

Lesson 2: Start Small

When Harting’s leadership set a goal to double turnover by 2030 with the same staff, Nattke knew speed was critical. The sales process’s biggest bottleneck was manual data entry from new orders. After testing three solutions, he chose Workist for its transparency and consistency. Orders now process four times faster. “In the beginning, errors will happen, so start small, learn, and then grow,” Nattke advises. He recommends beginning with an “error resistant” part of the business to build trust before scaling. Müller agrees: “Your first AI project really should work. You don’t want to burn AI for your entire organization.” Dungey urges manufacturers to start with a business problem rather than the technology, warning against hype. “The best gains come from modest, targeted interventions,” he says, not moonshots. But scale must be considered from the start to avoid “pilot purgatory,” where a system works perfectly in theory but can’t handle real-world messiness. Rajagopal adds her 80/20 rule: 80 percent of value comes from AI modeling specific to your company, and 20 percent from general productivity tasks. “Invest in both,” she advises.

Lesson 3: Don’t Underestimate Data Preparedness

Manufacturers like Siemens, Mercer, and Michelin have integrated AI into HR, safety, supply chain, and customer service, but none of this works without solid data. “None of this would have been possible if we had not put in a great deal of less glamorous effort on the data side,” says Rajagopal. Properly labeled and organized data is non-negotiable. Michelin assigned “data owners” to each business segment, who learned who needed what data and when, ensuring it stayed clean and accessible. “If you don’t give the right level of strategic importance to data governance, you will see the outcome in the rate of AI adoption,” she warns. On the factory floor, data management adds layers of complexity, from sensor connections to metadata. Dungey believes “perfect data” is impossible but insists it must be relevant, reliable, and governed properly. Edge cases are crucial because they stump AI, and systems should defer to humans when uncertain. Even external AI products like Workist require clean datasets. Nattke emphasizes, “You need to take the time in the beginning to train the system properly. If you think, ‘Oh, yeah. It’s AI. It knows everything.

It’ll do it right’,it won’t.” Müller sums it up: “Clean up your master data!”