This article builds on discussions from the “AI, Ethics and Manufacturing” event hosted by the Manufacturing Technical Network in collaboration with the Mersey & West Cheshire Local Network and the AI Technical Network, which took place on 27 January 2026. The session explored the practical realities of implementing AI in manufacturing – moving beyond ambition and into execution.
What became clear is that the challenge is no longer understanding AI – it’s making it work where it actually matters. The technology is advancing rapidly, and its potential is widely recognised. The real difficulty lies in applying it effectively within operational environments, where processes are complex, data is imperfect, and decisions carry real consequences.
In one manufacturing facility, a production line had been experiencing recurring, unplanned shutdowns. Each incident triggered a familiar cycle: manual inspection, reactive maintenance, and post-event analysis. The organisation was not short of data. In fact, it had years of it (sensors, control systems, historian logs), everything was there. What was missing was insight.
An AI-driven predictive maintenance model was eventually introduced. Within weeks, it began identifying subtle vibration patterns and thermal deviations that preceded failure events. These signals had existed in the data for years but gone unnoticed.
Downtime reduced, maintenance became proactive and decision-making improved. Yet, six months later the initiative stalled. Not because the technology failed, but because the organisation was not ready for it. Data pipelines were inconsistent. Engineering teams did not fully trust the outputs. Operational routines had not been adapted to act on AI-generated insights.
This is not an isolated case. It is happening across various industries, and it leads to a fundamental paradox.
The AI Paradox: High Investment, Limited Return
AI is often presented as the next frontier in manufacturing performance. According to McKinsey & Company, it has the potential to generate between $1.2 trillion and $2 trillion annually across manufacturing and supply chain operations (McKinsey Global Institute, 2021).
The opportunity is clear. The results, however, are far less consistent. This is reflected in industry data. A recent State of Smart Manufacturing report by Rockwell Automation (2025) highlights that while AI adoption is accelerating across manufacturing, many organisations are still struggling to translate investment into measurable operational outcomes.
Research from MIT Sloan Management Review and Boston Consulting Group shows that while AI adoption is increasing, only a small proportion of organisations achieve meaningful financial returns, with many initiatives failing to move beyond pilot stages.
In my experience, this is where many organisations go wrong. They assume the challenge is technological. It rarely is.
Manufacturing Is a Process Discipline – AI Is Not a Shortcut
Manufacturing has always been built on process discipline. Stability, repeatability, and control are what drive performance. This is not new – it is the foundation of Lean, Six Sigma, and every serious operational excellence programme.
AI does not change that. If anything, it reinforces it. Too often, organisations start with the question: “Where can we apply AI?”
It sounds logical, but it’s not.The better question is: “What operational problem are we trying to solve, and do we truly understand it?”
Without that clarity, AI becomes a solution in search of a problem.The result is predictable:
- Pilots with no clear business case
- Solutions disconnected from operations
- Investment without impact
AI does not fix broken processes. If anything, it makes their weaknesses impossible to ignore.
Where AI Actually Delivers Value
This brings us to a more useful question – one raised in recent industry discussions: “Is AI better at solving problems, or at making sense of complexity?”
In manufacturing, the answer is clear. AI creates the most value when it improves how decisions are made, not when it attempts to replace them. Its strength lies in:
- Identifying patterns in large, complex datasets
- Detecting anomalies in real time
- Supporting optimisation across multiple variables
But the decision still matters, and so does the engineer.
Case Example: Quality Variation in Production
In a high-volume manufacturing environment, a persistent quality issue had resisted traditional analysis. Process parameters appeared stable. Statistical tools showed no clear cause. An AI model was applied to historical production data. It identified a subtle interaction between ambient conditions and calibration drift, something no one had previously considered.
The fix was straightforward. The insight was not. The AI model identified a previously unseen relationship between environmental conditions and calibration drift, allowing engineers to isolate the root cause and adjust process controls accordingly. Defect rates dropped by over 30%. That is where AI proves its worth – not by replacing expertise, but by extending it.
From Industry 4.0 to 5.0: Re-Centring the Human
This shift in thinking aligns closely with the transition from Industry 4.0 to Industry 5.0. Industry 4.0 focused on automation and digitisation. The goal was often to reduce human intervention. Industry 5.0, as outlined by the European Commission, takes a different view; placing emphasis on human-centric systems, resilience, and sustainability (European Commission, 2021).
This distinction matters because the most effective AI-enabled organisations are not removing humans from the system, they are redefining their role. Engineers are no longer just executing processes. They are:
- Interpreting AI-generated insights
- Challenging outputs
- Making decisions in increasingly complex environments
This has implications for capability development. Entry-level engineering roles are already changing. Routine tasks (drafting, reporting, basic analysis) are increasingly automated. What replaces it is not a gap, but a shift. Graduates will need to operate at a higher level from day one:
- Managing AI-assisted workflows
- Applying judgement to machine-generated outputs. This creates a potential capability gap: without hands-on experience in producing these outputs, future engineers may be expected to validate results they no longer fully understand.
- Understanding systems, not just tasks
That transition will not be easy (particularly for those already in the middle of their careers), but it’s unavoidable.
The Scale-Up Problem: Why AI Stalls After the Pilot
Even when AI works technically, it often fails organisationally. Many companies can demonstrate a successful pilot. Far fewer can scale it. This is sometimes referred to as the “second valley of death”—a term used to describe the gap between innovation and successful implementation (Markham, 2002). In reality, it reflects deeper systemic issues.
- Process Instability: AI depends on consistent data. Unstable processes generate unreliable inputs. If a production system is in constant firefighting mode, AI will not stabilise it. It will simply reflect the chaos more accurately. Lean fundamentals still matter. There is no way around it.
- Fragmented Data: Despite years of digital investment, many organisations still operate with siloed systems. According to PwC (2022), over 60% of industrial companies identify data integration as a major barrier to AI adoption. If the data cannot be connected, the model cannot scale.
- Misaligned Expectations: AI is often expected to deliver immediate, transformative results. That is rarely how it works. Value tends to be incremental. It builds over time. Organisations that approach AI as a long-term capability (not a quick win) consistently outperform those that don’t.
- Capability Gaps: There is a growing need for individuals who can operate across disciplines – engineering, data, and operations. They are still relatively rare. Without them, AI remains either a technical exercise or a theoretical one. It does not become operational.
Ethics, Governance, and Responsibility
As AI becomes more embedded in manufacturing, the questions become broader. Not just “Does it work?” but:
- Is it fair?
- Is it transparent?
- Is it accountable?
The OECD AI Principles (2019) emphasise these issues, and rightly so, because AI will optimise for whatever it is designed to measure.
Case Example: Supply Chain Optimisation
In one global manufacturing network, an AI model was used to optimise sourcing decisions by analysing cost, lead time, and supplier performance across multiple regions. The model successfully reduced procurement costs and improved delivery reliability.
However, it also redirected production towards regions with lower environmental and labour standards, simply because those variables were not part of the optimisation logic. This is the risk. AI will optimise for what it is designed to measure. Ethical outcomes require deliberate design, not assumption.
The Environmental Trade-Off
There is another dimension that receives far less attention. AI requires infrastructure (data centres, computing power, and cooling systems) all of which consume energy and, in many cases, large volumes of water (Parliamentary Office of Science and Technology, 2024). Research by Strubell et al. (2019) highlighted the significant carbon footprint associated with training large AI models.
For manufacturing organisations already under pressure to reduce emissions, this creates a tension. We may be improving efficiency in one part of the system while increasing environmental cost in another. This is not an argument against AI, but rather an argument for using it responsibly.
Where to Start: Practical, Scalable Use Cases
For organisations still early in their journey, the priority should not be transformation. It should be traction. The most effective starting points are:
- Clearly defined
- Operationally relevant
- Measurable
Three areas consistently deliver value:
- Predictive Maintenance: Reduces unplanned downtime and builds on existing data.
- Computer Vision for Quality: Improves consistency and reduces subjectivity.
- Production Planning and Forecasting: Enhances scheduling and reduces waste.
None of these are revolutionary, and that’s precisely why they work.
A well-documented example comes from Siemens, where AI-driven predictive maintenance is used to monitor equipment performance and identify failure patterns in advance. By analysing sensor data across production systems, this approach has delivered measurable reductions in unplanned downtime and maintenance costs (Verysell AI, 2024).
The Future: Not Autonomous, But Integrated
Looking ahead, the conversation is shifting again. There is increasing interest in neuro-symbolic AI – systems that combine machine learning with rule-based logic. The goal is simple: more capability, more control, and more trust.Fully integrated systems are still evolving, but the direction is clear. AI in manufacturing will not be fully autonomous. It will be structured, governed, and human-led.
Conclusion: Engineering Value, Not Chasing Technology
AI is not a strategy. It is a capability, and its value depends entirely on how it is applied. The organisations that succeed will be those that treat AI not as a standalone initiative, but as part of a broader system of engineering excellence. They will start with real problems, build strong process foundations, develop the right capabilities, and integrate AI into how decisions are made every day.
Not as an add-on. Not as a pilot. But as part of how the organisation operates, because ultimately, AI is not about replacing engineers, it’s about enabling them to operate at a higher level – making better decisions, faster, with greater clarity.
The question is no longer whether AI will shape manufacturing. It already is. The real question is whether organisations are prepared to use it in a way that delivers meaningful, measurable, and sustainable value.
References
European Commission (2021) Industry 5.0: Towards a sustainable, human-centric and resilient European industry. Brussels: European Commission.
Markham, S.K. (2002) ‘Moving technologies from lab to market’, Research-Technology Management, 45(6), pp. 31–42.
McKinsey Global Institute (2021) The State of AI in 2021. Available at: https://www.mckinsey.com (Accessed: 17 March 2026).
OECD (2019) OECD Principles on Artificial Intelligence. Paris: Organisation for Economic Co-operation and Development.
Parliamentary Office of Science and Technology (2024) Artificial Intelligence: POSTnote 762. UK Parliament. Available at: https://post.parliament.uk/research-briefings/post-pn-0762/ (Accessed: 22 March 2026).
PwC (2022) Industrial Manufacturing Trends Report. Available at: https://www.pwc.com (Accessed: 17 March 2026).
Rockwell Automation (2025) State of Smart Manufacturing Report. Available at: https://www.rockwellautomation.com/en-us/capabilities/digital-transformation/state-of-smart-manufacturing.html (Accessed: 25 March 2026).
Strubell, E., Ganesh, A. and McCallum, A. (2019) ‘Energy and Policy Considerations for Deep Learning in NLP’, in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 3645–3650.
Verysell AI (2024) AI in Manufacturing: 5 Inspiring Real-World Success Stories. Available at: https://verysell.ai/ai-in-manufacturing-5-inspiring-real-world-success/ (Accessed: 25 March 2026).