The manufacturers succeeding with AI stopped treating it as an AI project
Manufacturers have spent the past three years experimenting with AI. Pilot projects have proliferated across factories, supply chains and engineering operations. Yet despite the excitement, many organizations remain stuck in the same position they occupied at the start of their AI journey: plenty of proof-of-concepts, very little operational impact.
The problem, according to Sunitha Rao, SVP/GM of Hybrid Cloud and AI Infrastructure at Hitachi Vantara, is that too many manufacturers still view AI as a technology initiative rather than an operational one. The organizations generating measurable value are not necessarily deploying more sophisticated models. They are embedding AI into the way the business already works.
“The first big-ticket item isn’t the AI capability,” Rao says. “It’s about how it is operationalized. The organizations creating value are moving from experimenting with models to industrializing them with governance, workflows and data foundations that can influence real-world decisions. The ones that remain stuck in experimentation are usually the ones that have gaps in how data flows through those operational processes.”
That distinction may explain why some manufacturers are moving rapidly from pilots to production while others continue to struggle. Success increasingly depends less on the sophistication of the technology and more on whether organizations can connect AI to measurable business outcomes.
Why pilots stay pilots
The manufacturing sector is hardly short of AI initiatives. Predictive maintenance, quality inspection, production optimization and supply chain planning all offer compelling opportunities to apply machine learning and analytics. Yet many projects never progress beyond the pilot stage.
The difference between successful and unsuccessful deployments often comes down to where AI sits within the organization. Manufacturers that treat it as a separate initiative frequently struggle to scale beyond experimentation, while those embedding it into operational processes are more likely to generate measurable returns. “I think the biggest inhibitor is treating AI as a standalone initiative,” Rao says. “It is not something you do as a separate transformation project. It is something you integrate into the systems and processes that already exist. The lesson manufacturers need to learn is how to connect data with outcomes and then embed AI into day-to-day operations.”
She points to Siemens as an example of an organization that has successfully moved beyond experimentation. Rather than focusing solely on AI models, the company integrated industrial data, digital twins and operational workflows into a broader operating framework. “It’s no longer about the models or the technical details of what AI needs,” Rao says. “It’s about creating a continuum that takes information from one end of the process to the other and delivers a measurable outcome. AI becomes part of that flow rather than something that sits alongside it.”
The manufacturers generating the strongest returns increasingly view AI as another operational capability rather than a separate technology program. That shift in mindset appears to be one of the defining differences between organizations scaling AI successfully and those remaining trapped in perpetual experimentation.
Trust matters more than data
Data quality is often cited as the biggest barrier to AI success. Rao believes the issue has evolved. “I don’t think it’s about data anymore,” she says. “The bigger challenge is trust, connectivity and contextualization. Those are the areas where manufacturers still have gaps.”
Collecting information has become relatively straightforward. Manufacturing environments generate vast quantities of operational, engineering and business data every day. The challenge is understanding how that information relates to operational outcomes and ensuring it can be trusted when decisions are being made. “The organizations seeing value from AI are not necessarily chasing every new model that appears,” Rao explains. “They are investing in governance, creating operational context and building data foundations that can be reused across multiple workflows. They are treating data as a business asset rather than a technical asset.”
Context has become particularly important. Information rarely exists in isolation. Production data influences quality outcomes. Quality data affects supply chain decisions. Supply chain performance shapes production planning. Manufacturers need to understand these relationships if they want AI to generate meaningful recommendations. “The big-ticket item is contextualization because it has a direct impact on everything you’re investing in across your AI workflows,” Rao says. “Without that context, you’re not building an operational foundation. You’re just collecting information.”
The emphasis on trust is also changing how manufacturers think about governance. Rather than viewing governance as a compliance exercise, leading organizations increasingly see it as an enabler of operational decision-making.
Building foundations for scale
If there is one theme that runs through almost every manufacturing AI discussion, it is integration. Legacy systems, operational technology environments, cloud platforms and enterprise applications were rarely designed to work together in the way modern AI initiatives require.
Rao identifies three recurring challenges: OT-to-IT connectivity, cross-functional data silos and the complexity of hybrid environments. “Manufacturers are operating across multiple layers of technology that were never designed to create a continuous flow of information,” she says. “You have operational systems, enterprise systems, cloud platforms and different data environments. The challenge is connecting all of those pieces together.”
Many enterprises initially assume the answer lies in replacing older systems. Rao argues that approach is often unnecessary. “The successful organizations are not carrying out large-scale replacements. They are creating unified data layers, common governance frameworks and architectures that allow information to move across environments consistently. The key lesson is that this cannot be a standalone initiative. It has to be a strategic data foundation that enables automation and real-time operational decision-making.”
The growing adoption of hybrid cloud strategies reflects this reality. Manufacturers increasingly recognize that different workloads belong in different environments. Some applications require low latency, high performance and strict security controls. Others benefit from the scalability and flexibility of cloud infrastructure. “Manufacturers are not choosing between cloud and on-premises environments anymore,” Rao says. “They’re adopting hybrid approaches. The goal is to place workloads in the environment that makes the most sense from a performance, security, compliance and cost perspective.”
The challenge is ensuring data remains connected regardless of where it resides. Without that continuity, AI initiatives quickly encounter the same barriers that have slowed previous digital transformation programs.
Turning insight into outcomes
Manufacturing has never suffered from a lack of information. The real challenge is transforming that information into operational advantage. “I think the manufacturers gaining the most value are not necessarily collecting more data,” Rao says. “They’re focusing on making data usable. They ask three questions: how trusted is the data, how connected is the data and how actionable is the data?”
Those questions increasingly determine whether AI delivers measurable value. Companies that can establish trusted, connected data foundations are able to move more quickly from insight to action. Rather than using analytics simply to monitor performance, they use it to influence outcomes. “We’re no longer talking about collecting an insight and asking, ‘so what?'” Rao says. “Data should become an operational advantage. It should help predict failures, improve quality, optimise production and support better business outcomes.”
The strongest use cases emerge in predictive maintenance, quality inspection, production optimization and supply chain planning. These are areas where AI can influence measurable operational metrics such as downtime, throughput, scrap rates and inventory efficiency.
Rao points to Hitachi Rail as an example. Predictive maintenance initiatives help reduce equipment failures, lower maintenance costs and improve operational efficiency through earlier intervention and more informed decision-making. “The reason these use cases are important is because they are measurable,” she says. “They reduce downtime, improve quality, increase throughput and help organizations respond more quickly when conditions change. They demonstrate the return on investment and they can be scaled.”
Trust remains a critical factor throughout this process. The most successful manufacturers are positioning AI as a recommendation engine rather than an autonomous decision-maker. “We talk about keeping the human in the loop,” Rao says. “AI helps people make better decisions. It allows engineers and operators to validate recommendations before action is taken. Over time, that builds confidence because people can see the results and understand why recommendations are being made.”
Looking ahead, Rao believes manufacturers must move beyond isolated projects and focus on building the foundations required for scale. “Over the next few years, organizations need to move beyond pilots and focus on trusted data, governance and hybrid infrastructure,” she concludes. “They need to align business teams, operational teams and IT teams around common objectives and measurable outcomes. AI has to become part of how the organization operates every day.”
Ultimately, she argues that the manufacturers creating the greatest value from AI will not be the ones deploying the most models. They will be the ones that stop treating AI as a technology project and start treating it as an operational capability embedded throughout the business.
