September 18, 2026
Santosh MishraAI in Manufacturing Needs a Foundation: How MES Has Been Building It for Years
“AI in manufacturing is not being built from scratch. For three to four years, MES have quietly structured, contextualised and preserved the shop floor data that AI models now depend on. The real constraint is not missing data; it is how little of an existing foundation manufacturers are actually putting to work.”
In Brief
- a) Manufacturers are scaling AI ambitions faster than their data readiness: most expect a large share of processes to be AI-supported within years, yet a significant share admit their operational data is not yet fit for that purpose.
- b) Manufacturing Execution Systems (MES) have spent the past three to four years building exactly the kind of structured, contextualised, real-time data AI models require; this is existing infrastructure, not a new capability to be built from scratch.
- c) Predictive maintenance is the clearest proof point: AI models trained on MES-class historical and real-time data are already delivering double-digit reductions in downtime and maintenance cost.
- d) The constraint on AI in manufacturing today is not data generation; it is the gap between MES adoption and full enterprise integration, and between AI ambition and live production deployment.
The AI Ambition - Data Readiness Gap
Across manufacturing, artificial intelligence has moved from a research initiative to a boardroom expectation. Manufacturers surveyed for Rockwell Automation’s 2026 report, Scaling MES Across the Enterprise, expect 42% of their processes to be AI-supported within the next year, rising to 54% by 2030, a rapid pace of ambition for a set of technologies that, in most plants, remain in pilot. [1]
The same study surfaces a more revealing number: 43% of manufacturers acknowledge they are not effectively using the data they already collect, described in the report as the foundation artificial intelligence requires. [1]
That framing is worth pausing on. It implies the foundation exists. It has existed, in most manufacturing environments, for years; not as a bespoke AI initiative, but as the Manufacturing Execution System (MES) layer already running on the shop floor. Ninety-three percent of manufacturers now operate an MES in at least one facility, per the same research. The gap holding back AI in manufacturing is less about generating new data and more about using data infrastructure that was, in most cases, already built for another purpose.
This is the argument worth making explicit: over the past three to four years, MES has quietly done the work AI now depends on: standardising, contextualising and centralising shop floor data. Understanding what that work actually involved is the starting point for any credible AI strategy in manufacturing.
What “AI-Ready” Data Actually Requires
Before crediting MES with building AI’s foundation, it is worth being precise about what that foundation needs to look like. Machine learning models whether trained for predictive maintenance, quality prediction or scheduling optimisation are not effective when fed raw, disconnected signals. They require,
- 1) Structure – Data organised against a consistent model of assets, work orders and process steps, not free-floating sensor values
- 2) Context – A defect or downtime event is meaningless to a model unless it is linked to the specific machine, shift, material batch and process parameters active at the time
- 3) Continuity – A usable training set spans months or years of consistent, comparable records, not a single pilot’s worth of data
- Genealogy – The ability to trace a finished part or batch back through every process step it passed through, which is what makes root-cause and quality-prediction models possible at all.
- Raw IoT and sensor telemetry can supply volume. On its own, it cannot supply structure, context, continuity or genealogy. Those four properties are what turn a stream of numbers into something a model can learn from, and building them at scale, across a live production environment, is a data-engineering problem most manufacturers solved before AI made it fashionable to talk about.
What MES Has Already Solved (2022-2026)
This is precisely the gap MES was designed to close and has been closing for longer than the current AI conversation. Over the past three to four years in particular, MES platforms have evolved well beyond their original role of monitoring work orders and shift output,
- 1) Contextualised historians – Modern MES continuously links machine, quality and process data to specific work orders, operators and material lots, rather than storing timestamps in isolation
- 2) Genealogy and traceability – Every serialised part or batch carries a structured record of the process steps, parameters and inspections it passed through: the exact data structure quality-prediction and root-cause models are trained on
- 3) Standardised integration – MES now sits as the connective layer between PLCs, historians, quality systems and ERP, replacing siloed, plant-specific data formats with a common structure
- 4) Continuous operational history – Because MES has been running in production for years at most sites, it already holds the multi-year, comparable dataset that AI training requires
This is not a hypothetical capability. The global MES market itself reflects how central this layer has become: valued at USD 14.82 billion in 2024, it is projected to reach USD 25.78 billion by 2030, growing at roughly 10% a year, driven in large part, by manufacturers extending MES specifically to support AI-enabled production management. MES did not set out to be an AI project. It became one by already doing the groundwork. [3]
Where AI Is Already Running on That Foundation
Predictive maintenance is the clearest example of AI built directly on this kind of MES-class data, and it is far enough along to have documented results rather than projections.
McKinsey’s analysis of manufacturing analytics found that predictive maintenance models trained on the historical machine performance data that MES and connected historians generate; typically reduce machine downtime by 30 to 50 percent and extend machine life by 20 to 40 percent. [4] A separate McKinsey study of digitally enabled maintenance and reliability programmes found comparable gains on the cost side: a 5 to 15 percent increase in asset availability alongside an 18 to 25 percent reduction in maintenance costs. [5]
Both results depend on the same requirement set out in the previous section, a long, structured, contextualised history of how a specific machine has actually behaved. That is not sensor data in isolation; it is sensor data reconciled against work orders, shift patterns and maintenance records, which is precisely the reconciliation MES already performs.
The same logic extends to quality prediction and yield optimisation, where models depend on genealogy-linked process and inspection data again, the structure MES was purpose-built to maintain; to trace a defect back to its originating process step rather than merely flagging that one occurred.
The Gap That Remains
None of this means AI in manufacturing is a solved problem, and a credible account of the opportunity has to say so plainly.
Gartner’s 2026 Market Guide for Manufacturing Execution Systems finds that AI deployment within MES itself is still at an early stage: only around a third of MES vendors can point to AI running in a live production environment, rather than a pilot or a roadmap slide. Buyers, Gartner notes, want AI capability but remain cautious about data privacy and governance, a caution that is reasonable given how sensitive contextualised production data can be. [2]
The Rockwell research points to the same gap from the manufacturer’s side: while 93 percent operate MES somewhere in their organisation, only 23 percent have it fully integrated across the enterprise. [1] A foundation that exists in one plant and not the next ten is not yet a foundation an enterprise-wide AI strategy can be built on.
The honest read, then, is not that MES has solved AI adoption. It is that MES has already solved the harder, less visible half of the problem; building structured, trustworthy operational data, while the scaling and integration work still ahead is squarely a leadership decision, not a technology gap.
What This Means for Technology Leadership
For CIOs, CTOs and operations leaders, this reframes the AI conversation from where do we get more data to where is the data we already have underused. A few implications follow directly,
- 1) Audit before investing – Before commissioning new sensor networks or AI pilots, establish where MES already holds structured, contextualised history for the assets in question; that data is usually more complete than teams assume
- 2) Prioritise by data maturity, not by ambition – Sequence AI use cases against where MES integration is already strongest, rather than starting with the highest-value use case regardless of data readiness
- 3) Close the enterprise integration gap deliberately – The 93/23 split between MES adoption and full integration is the single largest lever available to widen AI’s usable data foundation, and it is an integration programme, not a research project
- 4) Treat data governance as an AI enabler, not a blocker – The data-privacy caution Gartner identifies among MES buyers is legitimate, and addressing it early avoids it becoming the reason a promising pilot never scales [2]
Conclusion
The story of AI in manufacturing is often told as though intelligence arrives first and infrastructure follows. In practice, for most manufacturers, the sequence has run the other way. MES has spent the past several years doing unglamorous, necessary work; standardising, contextualising and preserving operational data well before AI was the reason anyone asked for it. The manufacturers seeing results from AI today are, in large part, the ones whose MES was already doing this work.
At Motherson Technology Services, this is the foundation we work with manufacturing clients to build and extend: standardising shop floor data across PLCs, quality systems and ERP, and structuring it into the kind of contextualised, genealogy-linked record that the AI use cases described here depend on. That is not a separate initiative from AI adoption; it is the groundwork that makes AI adoption possible at all.
The manufacturers that treat MES as this kind of foundation, rather than a reporting tool to be modernised later, are the ones best placed to move from AI pilots to AI at scale.
References
About the Author:
Santosh Mishra, an accomplished technology leader with over 20 years of experience, specializes in leveraging innovation to address key business challenges in the manufacturing sector. As the head of IoT and Automation at MTSL, he drives strategic initiatives for digital transformation and smart manufacturing. His expertise lies in enhancing operational efficiency, eliminating non-value-added processes, and delivering actionable insights through real-time data analytics. By integrating machines, PLCs, ERP, MES, IoT, and advanced automation, he has successfully scaled digital transformation initiatives from proof of concept to enterprise-wide deployments across discrete and process manufacturing industries worldwide.
