Trust Me, It’s Air-Gapped
February 12, 2026
Machine Learning in Manufacturing Delivers ROI: 5 Takeaways From Think!AI
March 4, 2026Trust Me, It’s Air-Gapped
February 12, 2026Machine Learning in Manufacturing Delivers ROI: 5 Takeaways From Think!AI
March 4, 2026AI in Manufacturing
AI in Manufacturing Must Raise the Floor, Not Just Chase the Ceiling
A manufacturing team thought they had solved a knowledge accessibility problem.
They digitized manuals, loaded decades of best practices onto smart tablets, and sent operators outside into the field with instant access to the information they needed to do their jobs. The technology worked. The content was right. Operators even said they liked it.
But then usage dropped.
“It’s cold up north,” Matt Marcotte, Director of AI & Digital at United States Steel, told an audience at the Think!AI Summit hosted by the Pittsburgh Technology Council. “They want to go back into the shop and get a cup of coffee while they’re looking through the (paper) manual. That cup of coffee was more important than the convenience of having that out in the field on their tablet.”
That captures a core challenge facing AI adoption in manufacturing: Better tools, like AI, don't automatically translate into better outcomes.
It's a theme that Marcotte and fellow panelists Carlonda Reilly, VP & CTO of Kennametal and Geff Wood, Sr. Director ITAS Operations Portfolios at Alcoa, revisited a few times during a 45-minute panel discussion in front of roughly 300 attendees at the annual Think!AI Summit.
Manufacturing Still Runs on People
Across plants and industries, Marcotte said he sees the pattern repeat. Manufacturing systems look modern or automated, but consistent performance still hinges on a small number of highly experienced people who are on-site.
"We're a long way from AI being better than our best person," he said. "But can we get those newer folks, those people who aren't our best people, up by raising the floor and then strategically shooting through the ceiling where those opportunities present themselves?"
Peer-level buy-in matters more than formal change management plans. Executive support helps secure funding, but it doesn’t make projects succeed.
“You get that one operator who buys in,” Marcotte said. “He’s got the respect on the floor. He carries it forward.”
"We're a long way from AI being better than our best person,"
Raising the Floor
Reilly, of Kennametal, said her vision is to eventually use AI to reduce the amount of trial and error from the machining process.
Even with advanced CAD/CAM systems, the work still depends on operator experience learned over time. If a machining shop has an airline customer, for example, the goal is to cut a piece of aluminum into the right shape while producing the smallest amount of scrap material.
“Instead of doing that trial and error on a machine, what if you could put all that input, all that knowledge, into CAD/CAM and get the part out right the first time?" she said. "That's the vision. We're still a far ways off from getting all that through, but we have the building blocks starting right now in terms of developing algorithms with real data."
The Foundation Matters
As AI systems pull data from across operations, Wood, of Alcoa, pointed to the security challenges of moving data across the network in new ways in the interest of enabling AI tools.
"We've got to be able to secure that data, and at the same time, we are de-siloing that data," he said.
But there's even more complexity to de-siloing, he said.
"As you start to give that data the context that your AI systems need, now you realize, 'Oh, our ERP system refers to asset hierarchy totally different than our process control systems.'"
The ERP system might refer to a pump as XYZ 4, but on the production side, he said, it might be known as "the high-speed pump for spent liquor" (a metal production byproduct).
"Even GenAI can't bring those things together," he said.
"So those are challenges, and we're trying to work hard to meet it and without the IT and OT integration to help secure that data, bringing in good data governance that goes with all of this, it just won't work. Cybersecurity and OT teams need to be part of AI efforts from the start."
"To put a new technology
on the plant floor, we really need to see cold, hard proven ROI."
– ANDREW DRAKE
Advanced Analytics Practice Director