But while manufacturers see AI as a real solution, most are still struggling to scale those efforts.  

Manufacturers are beginning their AI efforts in areas where the value is easier to measure. Predictive maintenance and quality control are leading use cases, in part because they rely on well-defined datasets and offer clear return on investment. 

“These are data-rich environments where they can demonstrate ROI,” Sabin says. 

But while those applications are a logical starting point, they are not enough on their own. 

“Pilot programs are a great way to learn, but totally insufficient,” he adds. “To fully deploy and leverage AI requires changing many things, not just a software installation.” 

AI depends on access to consistent, high-quality data across systems. But in many facilities, that data remains fragmented. 

“Many manufacturers are lagging because their data still sits in a number of legacy silos,” Sabin says. 

AI systems rely on historical data to predict future outcomes. Without reliable, connected data, those predictions become less useful. 

“If you’ve got poor data, it’s the old saying: garbage in, garbage out,” Sabin notes. 

Unlocking that data requires integration across systems such as ERP and MES, as well as changes to how information flows across the organization. 

Complete scaling also requires rethinking the systems that data flows through. 

Fully replacing legacy systems is often impractical due to cost and disruption. At the same time, simply layering AI on top of outdated systems can create new problems. 

Instead, Sabin says manufacturers should take a hybrid approach: incrementally modernizing systems while introducing AI where it can deliver immediate value. 

Ultimately, Sabin says, the companies that succeed will be those that move beyond experimentation and treat AI as a core part of how their operations run: using it not to replace people, but to amplify what they can do.