From Strategy To Shop Floor: A Practical Road Map For Manufacturing AI

Fabio Caversan is the Global CTO at Stefanini Group, leading the SAI suite to co-create AI solutions for digital journeys.

gettyManufacturers don’t need another reminder that artificial intelligence is “the future.” They need a practical path to move from pilots to plant-wide impact.

That gap is real and it’s measurable. The 2026 Industrial AI Readiness Survey found that 64% of industrial organizations plan to or are already leveraging AI for predictive maintenance, while only 7% have AI embedded in core processes. The implication is clear: In most factories, the limiting factor is not the model; it is the plant.

In the same set of references, the constraint shows up where it always does in manufacturing: data and integration. Rockwell’s “State of Smart Manufacturing 2026” finds that only 43% of manufacturing data is used effectively. According to the Manufacturing Leadership Council, data problems are the top obstacle to AI for nearly 70% of manufacturers. If your AI program is struggling to scale, odds are you’re not facing an “AI problem”; you’re facing modernization, data and execution problems.

As manufacturing leaders, we need to stop framing AI as a single initiative and start treating it as an operating model that can be summarized in five verbs: modernize the plant, connect its data, execute with context, act with AI and physical AI and protect it all.​

Most plants still run on a mix of legacy environments that were never designed for today’s pace of change. The challenge is not that these systems don’t work; it’s that they are difficult to evolve. Documentation is incomplete, tribal knowledge is concentrated in a few experts and migrations feel like high-risk events.

AI speeds up modernization by automating discovery, translation and validation. Rather than replacing engineers, it compresses migration timelines while safeguarding operational stability. The impact is already visible: An auto parts provider recently migrated from GE iFIX to Ignition SCADA on a tight two-month deadline using a small team.​

This matters because the market is moving. The North American industrial controls market is projected to grow from $50.26 billion in 2026 to $64.7 billion by 2031. Updating legacy code and control models is increasingly the first step toward an intelligent enterprise.

Manufacturing has never lacked data. It has lacked usable data.

Plants often have historians, OPC connectivity and thousands of tags streaming from equipment. Yet the meaning of that data varies by site, line and even shift. When every plant names assets differently and every system carries its own context, scaling analytics becomes a reinvention exercise.

Effective industrial data management is crucial for scaling AI. By treating data context as infrastructure—using governed plant models like ISA-95 and unified namespaces—organizations create a consistent language that makes integrations repeatable across facilities.​

Even the best insights are wasted if they don’t change what happens on the floor.

That’s why manufacturing execution matters. MES is where production reality is captured and managed: work orders, quality checks, traceability, downtime reasons, labor and material consumption. It is also where AI becomes measurable, because MES connects decisions to outcomes.

In many organizations, the MES conversation becomes polarized: packaged platform versus custom build. The better question is simpler: What does the plant need to execute reliably, and what constraints does the environment impose? The right answer varies by process, regulatory requirements, legacy footprint and the speed at which the business needs to standardize across sites.

A vendor-neutral design philosophy is essential. By evaluating options based on runtime performance, customization flexibility, PLC connectivity and long-term support, organizations can achieve extraordinary results.

Once plants modernize, connect and execute, AI can move from reporting to action.

This is where the next wave of manufacturing value emerges: systems that perceive, reason and act, often in closed-loop scenarios. In the summit framing, this includes AI-assisted systems and what many now call “physical AI”—AI that influences the physical world through automation, robotics and control. While pre-programmed robots still handle roughly 90% of routine manufacturing tasks, physical AI and camera-based closed-loop controls are taking process automation to the next level.

Consider a real-world application in the materials sector: A paint company facing frequent quality escapes used camera systems and AI control logic to monitor powder-cooling processes and automatically adjust belt speed or stop the line when necessary. This single intervention eliminated 33% of quality issues. Elsewhere, global fertilizer companies have leveraged autonomous indoor drones and digital-twin robot training to slash inventory inspection times from 24 hours to one hour, while pushing inspection accuracy to 98%.

The nuance is governance. Closed-loop does not mean uncontrolled. The most effective deployments are designed with clear boundaries, approvals and rollback paths because manufacturing is a high-consequence environment where stability is a feature, not a constraint.

​As manufacturers modernize controls, expose data to enterprise systems and deploy AI closer to operations, cybersecurity stops being a parallel program and becomes part of the architecture. The OT security market is projected to grow from $27.4 billion in 2026 to $58.9 billion by 2031. Yet operational preparedness lags behind: Only 45% of manufacturers feel adequately prepared for IT/OT convergence security, even as 80% report a rise in cyber incidents. And the business impact is not abstract: “62% of manufacturing firms experienced a resilience or availability issue, typically costing US$200,000 to US$2 million.”

OT environments have different constraints from those in IT environments—uptime is paramount, patching cycles are slower and legacy protocols are common. But the direction of travel is clear: IT/OT convergence is accelerating, and security must be designed into modernization, data and AI programs from day one.

The plant of the future is not built overnight. Industrial leaders succeed by starting small, then scaling. By aligning modernization, data management, production execution, physical AI and robust security into a unified strategy, manufacturers can finally move past the pilot phase and turn industrial AI into measurable, scalable growth.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?