The recently published eBook on Industry 4.0 and AI: What You Can Learn from the Leaders paints a revealing picture of where manufacturing stands today. Progress is real, investment is flowing, and ambition is high. Yet beneath the surface, a quieter crisis is brewing, one that won’t announce itself with a factory alarm, but will show up in boardroom disappointment.
W. Edwards Deming warned us decades ago: “The greatest danger in times of turbulence is not the turbulence itself; it is to act with yesterday’s logic.” Replace “logic” with “yesterday’s data and tools,” and you have a precise diagnosis of where many manufacturers find themselves in 2026.
Executives are not waiting
Let’s be direct. Boards and C-suites are not reading maturity assessments and benchmark reports with patience. They are reading AI headlines, watching competitors announce double-digit productivity gains, and asking one question: when does this show up in our numbers?
The survey findings are sobering in this context. Only half of manufacturers have a formal Industry 4.0 strategy. IT and OT teams are fully aligned at virtually no company surveyed. And meaningful AI in production operations? Still rare. The urgency has not diminished. If anything, the gap between executive expectation and operational reality is widening.
The turbulence Deming spoke of is here. The question is whether your organisation is responding with the tools and thinking of tomorrow, or defending itself with the infrastructure of the past.
The pace of change is accelerating
Before we talk about what manufacturers should do, it is worth pausing to feel the weight of how fast the external environment is moving.
In 2025, Anthropic released Claude 3.5 and subsequently Claude 3.7, models that demonstrated significant leaps in complex reasoning and coding capability within months of each other. Google launched Gemini 2.0 with native multimodal capabilities, processing text, images, audio, and structured data simultaneously, and followed it rapidly with further iterations. OpenAI continued its cadence of model releases, advancing agentic capabilities that allow AI systems to autonomously execute multi-step workflows with minimal human intervention. Meanwhile, Meta, Mistral, and a growing field of open-source contributors ensured that frontier AI capability was no longer the exclusive domain of a handful of technology giants.
Each of these releases represents expanding capability frontiers directly relevant to manufacturing: more accurate anomaly detection, more reliable process reasoning, and more capable autonomous agents that can operate within industrial environments. The models being released today are being trained on more data, with more sophisticated architectures, and with explicit focus on enterprise and industrial applications.
The implication for manufacturing leaders is stark: the capability gap between what AI can do and what your organisation is ready to use is not closing on its own. Every month that passes without a governed, structured data foundation in place is a month of compounding readiness debt. The technology will not wait for your data architecture to catch up. It will simply reward those who were ready first.
This is precisely why the urgency that drove the conversations at the 2025 MES & Industry 4.0 Summit has not faded. If anything, the acceleration of AI development makes the foundational work more urgent. The leaders in that room understood something that the broader manufacturing community is still absorbing: you cannot retrofit data quality onto an AI strategy. You have to build it first.
The inconvenient truth about data
The eBook is candid on this point: data approaches across manufacturers are fragmented, inconsistent, and in many cases still Excel-dependent. This matters enormously for AI. You might not be aware, but there is more fiction written in Excel than in Word!
There is a fundamental distinction that every executive must internalise: the difference between deterministic, structured data and non-deterministic, unstructured data, and the profound impact each has on what AI can and cannot do for you:
- Structured, deterministic data is precise, consistent, timestamped, and governed. It tells a machine, or an AI model, exactly what happened, when, under what conditions, and with what outcome. It is the language of reliable prediction, closed-loop quality control, and trustworthy automation.
- Non-deterministic, unstructured data, manual entries, freeform text, inconsistent formats, siloed systems speaking different languages, introduce noise at the input. And noise at the input does not stay contained. A model trained on poor data does not just return poor answers; it returns confident, poor answers, at scale, at speed.
The eBook highlights this precisely: manufacturers are dealing with structured production data alongside semi-structured engineering data, unstructured video and sensor feeds, and legacy inputs that were never designed to be machine-readable. Without deliberate architecture to govern, contextualise, and validate this data before it reaches an analytical or AI layer, the promise of intelligent operations remains exactly that, a promise.
Consider what this means in practice. An AI model asked to optimise production scheduling draws on equipment availability data, material lead times, operator capacity, and quality history. If any one of those inputs is inconsistently formatted, manually entered, or governed by different definitions across production sites, the model’s output is not just imprecise; it is potentially worse than the human decision it was meant to replace. The executive who approved the AI investment sees a failed pilot. The real culprit was never the AI. It was the data infrastructure that was never built to feed it.
The good news: the leaders are showing the way
The results are already in for those who moved early. Machine Learning is reducing defect rates in quality inspection. Real-time production monitoring is compressing decision cycles from hours to seconds. AI-powered process optimisation is delivering measurable yield improvements in semiconductor and electronics manufacturing. The technology works. The question was never whether AI could deliver value in manufacturing. The question is always whether your organisation is ready to receive it.
And that is where the most important leadership lesson emerges.
The manufacturers pulling ahead share one defining characteristic: they stopped debating the ROI of data infrastructure and started treating it as a non-negotiable corporate asset, the same way they treat energy, connectivity, or physical plant. No CFO demands a circuit-by-circuit ROI justification before wiring a new facility. The infrastructure is the precondition. Everything else runs on top of it. Deterministic, governed, structured data is no different.
One leading semiconductor manufacturer captured this precisely when challenged to justify the business case for a unified data architecture investment. The executive sponsor’s response was unambiguous: “We are not calculating the ROI of this. We are calculating the cost of not doing it.” That decision, made without a promised payback period, unlocked predictive quality models, AI-assisted scheduling, and yield loss response times measured in minutes rather than days. The return arrived. But only because the foundation was built first, on conviction rather than spreadsheet justification.
The companies stuck in pilot purgatory today are, almost without exception, the same companies still debating that foundation.
The encouraging reality is that the path forward is becoming less difficult to walk. MES platforms are maturing into genuine data backbones, creating, validating, and contextualising production data in real time. GenAI and agentic AI capabilities are being embedded directly into industrial software, meaning manufacturers will not build these capabilities from scratch. They will arrive inside the platforms already being deployed. The infrastructure investment made today becomes the competitive engine of tomorrow.
What must leadership do now?
The eBook’s findings point to three non-negotiable priorities for executives who are serious about closing the gap between ambition and outcome:
1. Treat data as a strategic asset, not an IT project
Data governance, contextualisation, and architecture decisions belong in the boardroom. The quality of your AI outputs is a direct function of the quality of your data inputs. This is a capital allocation decision. And given the pace at which AI capabilities are expanding, every quarter of delay in building that foundation is a quarter of competitive advantage surrendered.
2. Resolve the IT/OT divide; structurally, not symbolically
Joint goals, shared accountability, and unified data flows between information technology and operational technology are not optional enhancements. They are the foundational conditions for intelligent manufacturing. Committees and steering groups are not enough; structural integration is required.
3. Build foundations now, not after the AI strategy is ready
The manufacturers gaining ground are not waiting for their AI roadmap to be perfect before investing in MES, machine connectivity, and data infrastructure. They understand that the foundation enables the ambition, not the other way around. And with AI capabilities advancing at a pace the industry has never previously experienced, the window for building that foundation ahead of the curve is narrowing faster than most executive calendars appreciate.
The window is still open, but not indefinitely
The competitive landscape in manufacturing is not static. The companies building clean data architectures, deploying MES as their operational backbone, and piloting AI in controlled, high-value use cases today are creating a compounding advantage. Each year of structured data collection is a year of training signal that late movers will never recover.
And the pace of AI development means that the advantage is no longer linear. The manufacturers who are ready when the next wave of industrial AI capability arrives will be in a different competitive category entirely.
Deming was right. The danger is not the turbulence. The danger is arriving at the next strategic review cycle with the same data infrastructure, the same IT/OT silos, and the same pilot projects, while explaining to the board why the productivity curve has not yet bent.
The technology is ready. The use cases are proven. The leaders are demonstrating what is possible. The only remaining question is whether your organisation will act with tomorrow’s logic, or yesterday’s.
