Most mid-market manufacturers are stuck in AI pilot mode, and their data infrastructure is why

Not one mid-market manufacturer in Kaufman Rossin's latest research has reached what the firm defines as full AI operation, the stage where AI is simply how the business runs. That finding, drawn from a survey of senior decision-makers across U.S. mid-market companies, is the clearest signal yet that the manufacturing sector's AI moment is more complicated than the headlines suggest.

Writing in Automation World on July 8, 2026, Vera Nieuwland, director of Kaufman Rossin's business consulting services practice, lays out the numbers behind the gap. The data comes from Kaufman Rossin's report, "The State of AI in the Mid-Market," which covers industrial and mid-market manufacturers alongside other sectors.

The data infrastructure shortfall

The core problem isn't appetite. Some 91% of manufacturers surveyed plan to increase their AI investment, and every respondent agrees AI saves time. The problem is that AI requires clean, connected, and accessible data, and manufacturing has historically underinvested in exactly that.

Only 27% of manufacturing companies in the survey have a data warehouse or data lake, compared with 60% across the broader mid-market. Some 45% of manufacturers still operate with siloed data, and none reported using machine learning platforms. Across the entire mid-market, just 16% have reached a fully governed and integrated data state.

Legacy ERP systems compound the problem. Every manufacturer in the research runs on ERP, and those systems don't connect easily to modern AI tooling. Legacy integration ranks as the top AI barrier for manufacturers, cited by 55% of respondents, well above the 41% average across the broader mid-market.

Pilots are real, but they aren't scale

Nieuwland notes that the wins manufacturers have achieved so far are real but narrow: time savings in isolated workflows, accounts payable automation, individual productivity gains within processes that still span disconnected systems. These results matter, but they don't add up to transformation when the underlying data architecture hasn't changed.

The risk she identifies is treating a successful pilot as a finished journey. Experimentation registers as progress right up until it stalls, and organizational readiness is what bridges a promising proof of concept to operational scale.

There is also a cultural dimension that sits underneath the technical one. Industrial companies built competitive advantage on operational expertise and domain knowledge, not data-driven decision-making. Shifting to AI-driven operations asks those organizations to change a foundational assumption, one that decades of floor experience have reinforced.

Where to focus before the next pilot

Nieuwland outlines three priorities for manufacturers looking to move past the testing plateau. First, mapping existing data, knowing where it lives and how clean it is, and then breaking the silos that affect the highest-value workflows. One or two targeted integration platforms connecting the most-used systems can deliver meaningful progress without a full enterprise overhaul.

Second, she recommends identifying use cases where the data is already clean enough to demonstrate value at enterprise scale, then expanding outward from those wins rather than forcing AI onto fragmented data.

Third, and in her framing the hardest move: treating AI adoption as a culture shift rather than an IT project. Leadership has to position data as a strategic asset and model that orientation across functions. Tools don't change how organizations operate; people do.

What this means for your team

• Audit your data infrastructure before expanding AI pilots. Know which systems feed your highest-value processes and where the integration gaps actually are.
• Prioritize use cases where your data is already clean enough to demonstrate repeatable, enterprise-level results. Build outward from there rather than deploying broadly on fragmented data.
• Evaluate ERP integration capability explicitly when assessing any new AI tooling. The 55% barrier cited in the Kaufman Rossin research means this is where most deployments fail.
• Bring operations and IT leadership into the same room. The cultural barrier Nieuwland describes is an organizational design problem as much as a technology one.