From predictive maintenance to nearly fully automated factories, artificial intelligence and automation are drastically transforming how manufacturing plant managers operate.

One of the biggest changes these tools bring to manufacturing is the ability to move beyond just automation to provide greater access to expertise, said Jared Pfeiffer, principal client partner at Augury, who previously worked as a plant manager. 

Historically, manufacturing and most other industries’ operations relied on the knowledge of experienced workers. But as the older workforce retires and manufacturers face an acute labor shortage, AI can bridge the knowledge gap, Pfeiffer said.

“The opportunity with industrial AI is to combine that human experience with machine data and operational information so the best knowledge in the plant becomes available to everyone,” he said.

Experts say AI now can offer more than what even the most experienced and knowledgeable worker can provide.

“AI can begin bringing together sensor data, operating conditions, maintenance history, SOPs, manuals and institutional knowledge to help a less experienced employee understand what is happening and what they should work on,” Pfeiffer said. “For plant managers, that can mean shorter downtime events, faster troubleshooting, better shift-to-shift consistency and less dependence on a handful of experts.”

Because of AI’s advancing capabilities, rapid adoption has triggered concerns about workers being replaced. However, “AI is about supporting people, not removing them from the process,” said Dustin Katje, senior plant Manager at Pella Corp., a manufacturer with 20 locations. 

“It can reduce repetitive work and give team members more time to apply their experience, solve problems, and focus on work that creates greater value,” he said. “It allows our teams to spend more time driving improvements instead of mining the data to understand where we should point our resources.”

AI pilots vs. wide-scale adoption

A majority of manufacturers across the U.S. have experimented with AI and automation in some capacity, but not all of them have reached the stage of wide-scale adoption. Some are still exploring small pilot programs to gauge feasibility, potential use cases and return on investment. 

Katje said his company is doing both simultaneously, with permanent cross-functional teams working side by side.

“One group focuses on developing new applications, another on data and insights, and another on scaling proven solutions across our manufacturing network.” The goal, he said, is to “build solutions that can move across the company without being rebuilt each time.”

The Pella teams begin with problems that need to be solved immediately.

“We look at where our team members are losing time, where complexity creates opportunities for mistakes, or where better information could help us serve the customer,” Katje said. 

Pfeiffer is concerned that some companies move too quickly from “AI is exciting” to “we need AI everywhere” without clearly defining the problem they are trying to solve.

“The best business cases typically connect AI to existing plant metrics such as downtime, throughput, yield, scrap, maintenance cost, labor utilization, quality, or energy consumption,” he said.

Pfeiffer said the strongest ROI cases are tied to very specific operational problems. To prioritize these, he recommends company executives ask where their firm is losing the most production, what problems repeatedly require its best people, where operators spend unnecessary time searching for information and what failures or process deviations cost the most money. This information can be used to determine whether AI can help change any outcomes. 

Are plant managers ready for the changes?

While some manufacturers are celebrating significant productivity gains from AI, experts caution against moving too fast without fully safeguarding against emerging risks. 

“My biggest concern is AI companies that want to move aggressively into manufacturing because of the massive amounts of data the industry generates, but don't understand the safety and compliance requirements that come with it,” said Audra Kirkland, VP of manufacturing and IT at Hexion, a chemical manufacturer in Ohio.

In sensitive industries like chemicals, for instance, getting AI and automation processes wrong can create safety risks.

According to Ciarán O’Mara, co-founder and chief product and technology officer of Protex AI, the biggest change plant managers are seeing isn't the tech itself, but the expectation to become AI experts and embed it into their operations overnight. 

“It's a significant change for folks who have spent 20 or 30 years running lines with clipboards and stopwatches and aren't data scientists, but are now being asked to speak to LLM and agent deployment in front of their leadership,” O’Mara said. “It’s a context gap that needs to close, but one of the best ways plant managers can adapt is by adopting a ‘retrofit and not rebuild' mindset, leveraging tools available that work with the systems they currently have in place.”

This is a persistent issue manufacturers are still trying to solve. O’Mara said most of the industry is still in the pilot phase when it comes to implementing AI, and a lot of those pilots fail to produce clear returns on investment. According to a 2025 MIT report, 95% of pilot programs for enterprise AI solutions have shown no measurable impact. 

“Nearly 90% of AI projects don't survive their first year, and it's usually not because there's something wrong with the tech, but because nobody figured out how to fit AI into how the plant actually works,” O’Mara said. 

Trying to find a sustainable solution has led several companies to amend budgets and shift priorities. 

“Everyone assumes that companies will spend more on models and platforms, but we're actually seeing the rise in spend go toward the human capital that ensures AI is implemented correctly across the plant,” O’Mara said.

“That includes cleaning up and documenting siloed data across the organization, clear ownership and regulations, and keeping people accountable for the output because a great model on top of fragmented, undocumented data just gives you wrong answers faster.”