Why AI is not a low-hanging fruit for manufacturing, and what needs to change
Unless the manufacturing sector addresses several fundamental structural and operational challenges, the gap between AI being positioned as the magic solution to revolutionise factories and streamline production efficiencies, and its reality on the ground will only continue to grow.
From glossy trade show presentations to headline-grabbing media coverage, artificial intelligence (AI) is ubiquitous in any conversation, and certainly so in the manufacturing industry. However, the reality on the ground presents a markedly different narrative. Despite the excitement, AI in manufacturing remains far from being practical, scalable or transformative.
Industrial enterprises are generating more data than ever before. Forecasts suggest that, by 2030, manufacturers globally will produce a staggering 4.4 zettabytes of data – more than double the 1.9 zettabytes generated in 2023 (remember, one zettabyte is a trillion gigabytes). Discrete manufacturing alone is expected to account for 2.7 zettabytes, followed by automotive and process manufacturing.
On paper, that volume of data should fuel AI innovation. In practice, it is little more than digital noise.
Data – king or capital?
The core problem is not the lack of data – it’s the quality, context and usability of that data that matter.
Manufacturing data is messy, fragmented and operationally unstructured. It is real-time, transient, and often discarded as soon as the production task is complete. This situation contrasts with customer data in commercial sectors, where structured datasets fuel generative AI models for recommendation engines and chatbots. Operational data in manufacturing was never designed to persist or support future analytical purposes. It was created to control machines and processes – not to inform algorithms.
This is why, despite the vast amounts of data being generated, the industry struggles to move beyond pilot projects or one-off machine learning experiments. The foundations for widespread, repeatable AI use do not exist. The manufacturing sector is sitting on a digital goldmine but, without the tools to mine, refine, or extract value from it.
A common misconception driving AI conversations today is the belief that data is king. It is not. Data is capital – but like any form of capital, it has no intrinsic value unless it can be used.
Right now, much of the data in manufacturing environments is equivalent to having a suitcase full of foreign currency with no way to exchange it. The challenge is not about having more data, but about having the right data, in the right format, with the right framework in place to derive actionable insights.
Challenges in adopting AI in manufacturing
Industry conversations around AI adoption often fail to acknowledge this harsh reality. There is a tendency to focus on technology – on algorithms, software platforms, and shiny dashboards – without addressing the very real, and very practical, limitations that exist at the operational level.
Manufacturing data is often scattered across legacy systems, captured in different formats, missing critical context, and locked in departmental silos. Without overcoming these barriers, AI projects in manufacturing risk becoming expensive science experiments with little long-term value. Without clear problem statements or domain expertise, AI efforts often descend into exercises in proving technology without purpose. For example, a company collected large volumes of data only to conclude the obvious: that materials become more viscous when cold.
The sector’s experience with machine learning illustrates this point well. Machine learning models are already being deployed in manufacturing environments, but they are almost always bespoke, point-in-time solutions. They solve specific problems – such as predicting when a piece of equipment might fail, or identifying anomalies in production data – but they are rarely scalable or transferable across sites or use cases. Each implementation requires significant effort to clean, prepare and contextualise the data. And as soon as the problem changes, the solution has to be re-engineered.
This is why the vision of AI-powered factories running on autopilot is still a long way off.
Manufacturers have yet to establish the digital foundations needed to move from experimental, one-off projects to sustainable, scalable AI adoption. And this is not a technology problem, it is an operational, cultural and strategic one.
Mindset shift from data-first to value-first is a must
The starting point must be a fundamental shift in how manufacturers think about data and digital transformation. Treating operational data like a box of LEGO bricks is a useful way to understand the challenge. Manufacturers often spend months trying to sort every data point – every “brick” – before knowing what they’re trying to build. Instead, the focus should be on defining the desired outcome first, then organising only the relevant data to support it. This shift in mindset – from data-first to value-first – is where many companies go wrong.
The industry needs to stop chasing the technology headlines and start focusing on the basics. That means clearly defining business problems first, rather than leading with a solution looking for a problem. It requires organisations to develop frameworks that connect operational data to business outcomes and to build the context around the data so that it becomes usable.
It also means recognising that AI adoption is not a one-department job. It is a team sport. Manufacturing businesses need to break down organisational silos and foster collaboration between plant operators, IT teams, engineers, data scientists, and business leaders. AI will fail if it is left to a technical team working in isolation, disconnected from the people who understand the day-to-day realities of operations.
AI adoption succeeds only when it draws on the expertise of operators, engineers, and maintenance teams – the people who deeply understand the process environment. AI should put power into the hands of those with the problem, not just data scientists designing the model.
There is a need to be brutally honest about what is possible now, and what is not. While generative AI, including large language models, is still in the early stages of adoption across manufacturing, in targeted use cases such as predictive maintenance, generative design, and knowledge capture, broader integration is expected to accelerate significantly over the next 3-5 years. In fact, companies like Schneider, Siemens, and AVEVA are already exploring generative AI in OT environments. For example, operators are being supported by GenAI-powered assistants that interpret alarms, reference asset history, suggest fixes, and even generate work instructions – replicating the intuition of experienced staff in systems built for repeatable scale.
But while data challenges remain great, and business cases too fragile, it’s seemingly tough to justify the cost and complexity of these systems today.
The practical, proven AI applications in manufacturing today are in machine learning. Vision systems, predictive maintenance, quality assurance - these are the areas where machine learning models are delivering real value. But even here, the barriers to scaling are significant. Without robust data governance, clear business cases, and a framework that aligns digital initiatives to operational objectives, manufacturers risk getting stuck in an endless cycle of pilot projects that never translate into lasting change.
This is why so many AI initiatives in manufacturing fail to move beyond the proof-of-concept stage. The sector has spent years talking about digital transformation, but too often, that conversation has focused on technology at the expense of strategy, structure and culture.
For manufacturers that want to be AI-ready, the path forward is clear but challenging. It starts with aligning people, processes and technology. It means investing in data quality, breaking down silos, and embedding AI-readiness into digital strategies.
It’s also important to remember that AI is not just a technical enabler – it is a cultural disruptor. If organisations fail to align leadership, processes, and change management efforts, even the most advanced AI solutions can fail on day one.
AI is no different to when SCADA systems replaced analogue controls – it changes how people work, think and behave, and that shift has to be managed proactively and constructively. And perhaps, most importantly, it requires moving beyond the hype, and focusing on practical, achievable outcomes.
The manufacturing sector has the data it needs and even the technology. In most cases, it simply lacks the operational readiness to make it all work.
The potential of AI in manufacturing is real, but the road to get there is longer than many would like to admit. Closing that gap will require more than new algorithms or platforms. It will require new ways of thinking.
Until manufacturers recognise that, AI in manufacturing will remain an ambition – not a reality.