MES vs ERP vs SCADA: Where a Manufacturing Execution System Fits (ISA-95 Explained)

Almost every argument about MES is really an argument about boundaries. Somebody asks why the ERP cannot just do it, somebody else points out that SCADA already shows the machine data, and the discussion goes in circles because the three systems are being compared as if they competed. They do not. They sit at different levels and own different decisions.

ISA-95, the international standard for integrating enterprise and control systems, settles this cleanly. It is worth knowing not as theory but because it tells you which system should hold which piece of data — and that is the question that decides whether an integration works.

The four levels, in plain language

Level 1 is sensors and actuators — the physical signal. Level 2 is control: PLCs and SCADA, running the machine in real time, measured in milliseconds. Level 3 is manufacturing operations management: MES, working in minutes and hours, answering what is being made, on which machine, by whom, from which material, and at what quality. Level 4 is business planning: ERP, working in days and weeks, answering what to make, what to buy, what to invoice.

Each level talks to its neighbours, not across the stack. That is the whole point. An ERP reaching directly into a PLC produces exactly the brittleness the standard exists to prevent.

Why ERP cannot simply do MES

It is not a capability argument, it is a time-base argument. ERP transactions are designed to be correct and auditable, not fast and continuous. A work order in ERP is a commitment. A work order on the floor is a live thing that splits, pauses, changes machine, scraps four parts and resumes.

Push that reality into ERP and one of two things happens. Either the floor updates ERP in real time and the system becomes a data-entry burden that people batch up until the end of shift — which is how you get an ERP that is confidently eight hours wrong. Or the floor does not update it, and ERP plans against a stock figure and a capacity figure that were true yesterday.

MES exists to absorb that volatility, settle it, and hand ERP a clean confirmation: this order consumed this much, produced this much, took this long, at this quality. ERP gets accuracy without being asked to run at shop-floor speed.

Why SCADA is not MES either

SCADA sees more than MES does and remembers less that matters. It is built to visualise and control process state right now, at very high resolution, usually per line or per area.

What it does not carry is context. SCADA knows a machine ran at a given rate for forty minutes. It does not inherently know that those forty minutes belonged to work order 4471 for a customer with a delivery date, running material from a batch that also went into three other orders, operated by someone certified for that process. Genealogy, order context, labour, quality disposition and traceability are MES concerns.

This is why "we already have SCADA" is not an argument against MES, and why the reverse — replacing SCADA with MES — is a mistake too. They are not substitutes.

The eleven functions MES is actually expected to cover

The MESA model is the usual reference, and it is a useful buying checklist: resource allocation and status, operations scheduling detail, dispatching production units, document control, data collection and acquisition, labour management, quality management, process management, maintenance management, product tracking and genealogy, and performance analysis.

No plant needs all eleven on day one. But knowing the list stops you buying an OEE dashboard, calling it MES, and discovering a year later that there is no genealogy when a customer asks which batches a suspect lot went into.

What "MES automation" means once the boundaries are right

With levels respected, automation stops meaning dashboards and starts meaning decisions taken without a person.

A work order releases to a station only when material is reserved and the operator holds the right certification. A quality failure blocks the next operation instead of being caught at final inspection. A downtime reason raises a maintenance notification with the machine, the time and the fault already filled in. Consumption posts to ERP as it happens rather than as a shift-end guess.

Each of those is a loop closed inside level 3. None of them require ERP to get faster or SCADA to get smarter.

Digital factory software: where the term is useful and where it hides things

Digital factory is a portfolio word. It covers MES, APS, WMS, quality, maintenance, energy and analytics, and it is genuinely useful when a group needs one contract and one data model rather than seven integrations.

It hides things when it is used to avoid saying which of the eleven functions are actually implemented. The question that cuts through it: which system holds the work order, and which system holds material genealogy? If the answer is the same system for both, the integration is likely to hold. If they are different systems, ask who reconciles them.

Building the ROI case with numbers you already have

MES business cases fail when they are built on industry averages. They hold when they are built on four numbers the plant already knows, even roughly.

Unplanned downtime hours per month on the constraint, and the contribution margin of what that constraint produces per hour. Scrap and rework as a share of output, split between detected-late and detected-at-source. Hours per week spent producing reports by hand — count the shift handover, the customer report and the month-end reconciliation. And expedited freight plus emergency purchasing over the last twelve months, which is the price of not knowing.

Those four cover most of the realised return, and unlike a benchmark they can be argued about with the people who own them. A case built this way survives the CFO. A case built on a vendor’s average uplift figure does not.