Manufacturing execution systems are evolving from passive record-keeping to active, real-time decision support that shapes outcomes at the speed of operations. Senior leaders who still view MES as a historical ledger are missing the strategic pivot that modern manufacturing execution systems now demand.
The term MES emerged in the early 1990s, and for a long time, the job description appeared straightforward: track, trace, record, and report. That origin story still matters because many plants continue to treat the system as a compliant historian rather than the operational brain that modern volatility requires. Order flows in; work instructions flow out; a report lands days later. Meanwhile, supply shocks, demand swings, staffing gaps, and unplanned downtime now unfold on minute-by-minute timescales. The cadence of decision-making has outpaced the cadence of traditional record-keeping.
“Industry 4.0 raises expectations for value from connected factories, but none of the shiny technologies deliver much without robust execution,” Jeff Winter, Vice President Business Strategy at Critical Manufacturing, explains. “When MES is weak, AI has no context, digital twins lack feedback loops, and a cloud programme degenerates into a file cabinet in the sky. The organisations that scale digital transformation do not stumble into that; they modernise execution so the rest of the stack can work.”
That reframing changes how leaders evaluate return. A system that only reconciles yesterday’s scrap tells you what it cost; a system that spots an excursion as it emerges can prevent the loss in the first place. The difference is not cosmetic; it is architectural.
Why data architecture now decides MES impact
Real-time MES depends on data that is connected, contextualised, and immediately actionable. Streaming sensor values, human inputs, equipment states, material genealogy, and quality outcomes need to reside within a coherent model that business logic can traverse without incurring a translation tax. The lesson is unglamorous but decisive: without an agreed data language, analytics lives on an island, and execution cannot learn.
“The strategic leap happens when execution sits on a proper data foundation rather than treating data as an afterthought,” Winter continues. “A canonical model makes shop-floor context natively available, which means analytics are not bolted on as a report factory. Insights can route directly back into workflows, routings, and machine parameters. It stops being ‘analyse and admire’ and becomes ‘detect and act’ in the same environment.”
That design choice influences compute and infrastructure decisions. The streaming of telemetry across tools and lines is a bandwidth and integration challenge more than a heavy compute problem. The expensive cycles begin when you apply online analytics or machine learning to those streams and push decisions back into execution. Clarity about where inference lives, on the line, at the edge, or centrally, prevents performance theatre in which models exist but cannot meet latency budgets.
Integration without translation: the only way to enterprise speed
Manufacturing depends on interlocks. Enterprise resource planning systems release orders and materials; warehouse systems govern movement; laboratory and quality platforms certify IoT health; asset systems schedule maintenance. A responsive factory does not tolerate message-passing friction or semantic misunderstandings across these layers. Integration is more than APIs; it is a shared understanding of meaning.
“Connectivity is the easy part now; modern systems expose clean interfaces,” Winter adds. “The friction lives in context. If the downstream system does not understand the upstream data model, you do not have integration; you have relabelling. Successful plants insist on a single source of truth for execution-grade data, and they let every other system subscribe to that truth instead of reinventing it.”
This is also the reason modern MES programmes feel larger than a single application deployment. In many facilities, execution upgrades coincide with the retirement of legacy quality, analytics, or dispatch tools. The initiative stops being an integration project and becomes a replacement of functional systems. Stakeholders then face the fundamental question at the heart of flexibility: do you change the tool to match the process, or change the process to match the tool?
“Every manufacturer must decide how far to configure and how far to customise,” Winter explains. “Some will absorb process compromises to use standard modules for supportability. Others will fund deeper tailoring to preserve competitive know-how. There is no universal answer, but pretending your process matches an off-the-shelf template when it does not is how projects drift into brittle one-offs.”
Culture decides whether real-time MES creates value
Technology rarely fails in isolation; it fails when people use it the old way. The executive task is to align roles, incentives, and interfaces with the new mission. Operators will not adopt systems that bury critical information; engineers will not sustain models that cannot be interrogated; managers will not trust recommendations that cannot be explained. User experience and explainability are not nice-to-haves once the software decides set-points or dispatch logic.
“Plants with decades of MES experience need persuasion of a different kind,” Winter continues. “The job is not convincing them to adopt execution software; it is convincing them that yesterday’s execution approach cannot capitalise on today’s technologies. The argument lands when teams see that real-time MES is less about more screens and more about fewer, better decisions that happen in time to matter.”
Governance plays a surprising enabling role here. When data is properly labelled for sensitivity, lineage is visible, and quality states are clearly defined, teams can self-serve with confidence. Good governance is speed. It removes ambiguity in the path to production, rather than adding ceremony at the end.
Where decision rights should live
Debate continues about how far MES should extend into control. Standards offer a useful north star. ISA-95 establishes clear distinctions across levels, with PLCs at Level 1, SCADA at Level 2, MES at Level 3, and ERP at Level 4. Execution orchestrates the “what” and “when” across the plant; controllers and SCADA implement the “how” at deterministic speed. Confusing those layers blurs accountability and risks latency where none can be tolerated.
“If by ‘control’ someone means replacing PLC logic, that is the wrong ambition for MES,” Winter concludes. “If they mean orchestrating the shop floor with authority, sequencing, dispatching, gating quality, and coordinating resources, that has always been the remit. The trick is keeping the boundary crisp so decisions occur at the right level with the right guarantees.”
The implication is practical. Utilise MES to assess trade-offs among orders, resources, and constraints; leverage the automation layer to ensure safe, deterministic actuation. Let each domain excel.
What should real-time actually deliver?
Real-time without consequence is theatre. Value arrives when visibility triggers action quickly enough to change the outcome. That can mean pausing a tool before it runs into a known drift pattern detected in its telemetry. It can mean rerouting a lot when an upstream test indicates a marginal condition that a different path can remediate. It can mean rebalancing dispatch queues when a quality gate begins to slow throughput. The standard requirement is two-way connectivity between insight and execution.
Winter is explicit about the missing ingredient he encounters most often. “Enterprises collect oceans of data, but they cannot use much of it in the moment because it is not contextualised,” he continues. “A canonical model that merges equipment states, material identities, recipes, and quality outcomes is dull engineering work, but it is what makes prediction trustworthy and action immediate. Without that, you only report with more dashboards. You do not decide faster.”
The flexibility paradox and how to resolve it
Manufacturers want the speed of standard software and the differentiation of unique processes. Those desires collide when teams customise deeply and then struggle to maintain upgrades or expand to a second site with different constraints. There is a way out of the trap if leaders separate the layers. Encode true competitive advantage in well-defined, modular extensions; accept standard approaches for commodity functions; and enforce a strong domain model so that bespoke logic plugs in cleanly, rather than requiring rewiring of the core.
The rollout model should also reflect reality. Global programmes with ten plants rarely reduce to a single template. Treat them as ten projects that share a platform, a model, and integration patterns, while allowing site-specific flows to be configured rather than hard-coded. That balance is how you maintain velocity without diluting the edge that makes your process distinct.
Agents at the edge of what comes next
Generative systems enchant audiences with fluent text and images; the next phase is more operational. Agentic architectures combine reasoning with tool-use, allowing software to pursue goals through sequences of actions, not just produce answers. The relevance to execution is evident. An agent that can query the data platform, call forecasting or optimisation services, and then propose or implement a change to a routing or a set-point is more than a chatbot. It is a junior planner that never sleeps.
“The industry is crossing from ‘smart’ factories into ‘thinking’ factories,” Winter reflects. “The destination is autonomy, where routine trade-offs between quality, yield, cycle time, and cost are resolved in real time by systems. Agentic AI is a credible path because it turns generative ability into executable plans. The technology is here; the work now is designing workflows and controls, so agents operate safely within the rules of the process.”
None of that is feasible if the foundation remains a patchwork of isolated tables and ad-hoc scripts. Agentic systems need two guarantees: access to complete, contextualised information, and the ability to act through clean, governed interfaces. Those preconditions appear suspiciously similar to the exact prerequisites that modern MES already needs to succeed, which is precisely the point.
What will define a modern MES in five years?
The winners will present platforms more than monoliths. They will expose a canonical operational model of the plant, ingest high-rate streams and human context into that model, apply analytics where they make sense, and close the loop by feeding actions back into execution through safe, tested pathways. They will respect standards at the automation boundary and exploit openness to interoperate with ERP, quality, laboratory, warehouse, and asset systems without endless translation. They will be designed for change, because change is the only constant left.
Winter’s closing argument is deceptively simple. “Whatever arrives after agents will demand the same two capabilities: understand the state of operations in context, and influence that state with confidence,” he concludes. “Teams that invest now in those fundamentals will be ready for whatever the next wave brings. Teams that postpone the groundwork will keep buying new tools and discovering they cannot use them when it matters.”
The journey from record keeper to real-time strategist is not a slogan; it is a system design and an operating philosophy. Treat MES as the factory’s nervous system, sensing, interpreting, and coordinating across the enterprise, and the rest of the digital ambitions start to make sense. Treat it as a warehouse for yesterday’s events, and the gap between promise and practice will stay stubbornly wide.
