Manufacturing leaders face growing pressure to invest in artificial intelligence and demonstrate those investments can deliver measurable business value.

Smart Industry recently reported that 81% of more than 500 executives surveyed by PwC plan to boost AI spending over the next three years. That capital commitment forces a critical question for the C-suite: Is the next AI project advancing the organization toward a scalable strategy, or is it simply responding to the pressure to do something with AI?

See also: AI has a trust—not a technology—problem

For many manufacturers, especially those operating brownfield facilities, the bigger challenge is not finding a potential AI application. It is creating the operational foundation required for that application to produce reliable results and eventually scale.

Before investing in the technology, leadership should first define the desired business outcome. What problem are we trying to solve? Is AI the right technology for that problem? And if it is, how does that investment fit into the longer-term strategy for the plant?

Those questions quickly lead to an issue that historically has received much less attention in the boardroom: operational technology.

AI strategy requires an OT strategy

An effective manufacturing AI strategy cannot be separated from an OT strategy. AI readiness begins with connectivity. Equipment must be networked so that the necessary operational information is available.

But connecting machines alone is not enough. The data coming from production systems also must be cataloged, contextualized and normalized so that actions and process states carry consistent meaning.

Consider two machines performing related functions on the same production line. Experienced operators may understand that different tags, naming conventions or signals represent essentially the same event. An AI application does not inherently have that institutional knowledge.

See also: Physical AI: Where opportunity really sits—now, today—for manufacturers

If one machine identifies an action one way and the next machine describes the same action differently, the system has to be given enough context to understand that relationship. That is why collecting plant-floor data is not the same as making that data usable for AI.

This becomes especially challenging in brownfield environments, where plants may contain decades of equipment from different manufacturers, multiple generations of control systems and varying levels of automation. Very few facilities were designed from the beginning around a single data model or network architecture.

That does not mean manufacturers must wait until every asset is modernized before they can use AI. It does mean leadership needs a clear picture of what is ready today and what must change before AI can move further into the operation.

Start with the right problem, not the biggest problem

There are valuable applications for AI in manufacturing today. The key is matching the ambition of the application to the readiness of the environment.

Some of the strongest early opportunities can be deliberately narrow: analyzing large volumes of operational information to identify patterns, supporting business intelligence, recommending improvements or applying AI within a well-understood island of automation.

Expecting AI to control machines or manage processes across an entire production environment is a much greater leap.

A good example of the more targeted approach comes from a process chemistry application that E Tech Group developed to help scientists use machine learning to optimize experimentation.

SCADA-integrated dashboards built in ignition connected laboratory automation equipment, analytical instruments and cloud-hosted machine-learning models within a centralized environment.

See also: As AI adoption matures, data center controversies move to the fore

Scientists could manage experiment campaigns, monitor model-recommended experiments and capture results with much less manual tracking. The solution improved traceability and repeatability and established a foundation for future closed-loop autonomous experimentation.

The significance is not that every manufacturer should pursue this particular application. It is that AI was applied to a clearly defined problem within an environment where the necessary connectivity, data and process context could be established.

Manufacturers can take the same approach on the plant floor: Identify where the infrastructure and data already support a valuable use case, establish success there and continue building the foundation needed for broader applications.

IT practices cannot simply be transferred to OT

Building an AI-ready foundation requires another important distinction: IT and OT environments are not interchangeable. Networking principles may overlap, but OT networks support physical production processes where availability, predictable communications and production continuity create very different operating requirements.

An IT-oriented monitoring tool, for example, may actively scan connected devices and request information throughout a network. Applying that same methodology to an OT environment without understanding its architecture can introduce traffic into networks already carrying time-sensitive production communications.

See also: Successful AI products win long after the sales contract is signed

E Tech Group has encountered situations in which monitoring approaches brought into the plant from the IT environment contributed to intermittent latency and production problems. The technology itself was not necessarily the problem; the issue was applying an IT methodology without accounting for its downstream effect on OT operations.

IT and OT teams need to work together as connectivity expands. The principles may be similar, but execution in a production environment requires OT experience—particularly when changes must be made while production equipment continues to run.

Cybersecurity further raises the stakes. At one food-and-beverage facility, an earlier IT/OT assessment had identified network segmentation as a priority. When an enterprise cyber incident later occurred, that segmentation helped isolate production and allowed manufacturing operations to continue.

See also: When ransomware hits the factory floor

The lesson extends beyond cybersecurity. Work that may once have looked like back-end infrastructure—network segmentation, firewalls, OT architecture and asset visibility—now directly affects a manufacturer’s ability to support broader digital and AI initiatives.

A successful pilot does not prove you can scale

Manufacturing AI adoption is moving quickly, but adoption and maturity are not the same thing. Smart Industry also reported that a Rootstock survey of 520 digital-transformation leaders found roughly 94% of their manufacturing organizations were using some form of AI, even as much implementation remained in experimentation or pilot stages.

That distinction matters. A successful AI pilot proves a use case. It does not prove that the plant is ready to scale it. One of the biggest challenges in moving from pilot to production is accounting for edge cases.

AI can be extraordinarily powerful while still missing contextual conditions that appear obvious to an experienced human operator. People spend years learning what “normal” looks like in a production environment. They notice abnormalities almost instinctively.

An AI model only knows the conditions and context it has been given. A pilot therefore operates within a relatively constrained set of circumstances. As the application expands across a larger process, additional lines or multiple facilities, it will encounter conditions it has never seen before. Manufacturers should expect that.

An unexpected result after expansion should not automatically be interpreted as proof that the AI application failed. The model may need additional context, training or refinement as it encounters a broader range of operating conditions.

See also: Why IT/OT initiatives fail when executive engagement stops at sponsorship

Setting that expectation is important because a poorly prepared first experience with AI can create consequences beyond the individual application. A visible failure can undermine confidence among operators and leadership and make the organization reluctant to support the next initiative.

A well-selected application supported by the right infrastructure can have the opposite effect: it creates success the organization can learn from and build upon.

AI readiness Is a multiyear investment strategy

For executives deciding where to invest next, view AI as part of a broader operational roadmap rather than an isolated technology initiative. That roadmap begins with the intended business result. Leadership should be able to answer:

• What outcome are we buying?
• Why is AI appropriate for this problem?
• What needs to be true operationally for the application to work?
• Does this investment move the plant toward the architecture the organization ultimately wants?

Without that clarity, AI can create an unusually difficult capital allocation problem. When both the problem and potential solution remain vague, executives struggle to know what they are writing a check for or what measurable result to expect in return.

The same discipline should be applied to OT investments. For years, many manufacturing organizations have developed capital plans around visible production assets: This year a vision system, next year end-of-line packaging, later a PLC upgrade.

The less visible OT infrastructure behind those systems—including network architecture, firewalls, connectivity, data standards, and alarming—has not always received the same strategic attention.

AI changes the equation. Those once-overlooked details increasingly determine whether an organization can connect information across equipment, provide applications with meaningful context and eventually scale intelligence across production.

See also: Where AI belongs in OT and why secure integration matters more than speed

For a brownfield manufacturer, addressing that foundation may require a five- or 10-year roadmap. The objective should not be to modernize everything at once.

Leadership needs to determine what should be remediated this year, what comes next, which investments will yield immediate opportunities and how each project contributes to the target architecture. Planning a multiyear roadmap allows manufacturers to pursue AI where the environment is ready while continuing to modernize the rest of the operation.

The growing complexity may also change how manufacturers think about outside expertise. Maintaining deep internal capabilities across automation, networking, cybersecurity, data and AI disciplines will be difficult for many organizations.

Over time, manufacturers may rely less on partners for isolated projects and more on enduring relationships with organizations that understand both the plant environment and its long-term modernization strategy.

Build the foundation now to move faster later

The competitive question for manufacturers may ultimately be less about who launches the most AI pilots first and more about who creates the operational foundation that allows successful applications to scale.

There will always be pressure to pursue the newest AI opportunity. But an isolated application can become another technology island if the infrastructure around it is not moving toward a larger strategy.

See also: New Cisco AI study sees widening execution gap, strain on manufacturing infrastructure

Manufacturers that invest now in connectivity, OT architecture, cybersecurity and normalized operational data can still pursue targeted AI opportunities today. More importantly, each of those investments can make the next application easier to deploy.

Over time, that creates a compounding advantage: Each foundational investment can make subsequent AI applications easier to deploy and scale.

The manufacturer that builds one impressive AI application without addressing its broader OT environment may still have one application several years from now. The organization systematically building the foundation underneath AI may be positioned to move from one successful application to many.

For the C-suite, the more important AI strategy is not simply deciding where artificial intelligence can be used today but building an operational environment capable of supporting where the organization wants to go next.