An MES, or manufacturing execution system, is software that manages, monitors, and documents production on the plant floor in real time. It takes the orders your business system releases and executes them: dispatching work to lines, tracking material and labor, recording quality data, and building the genealogy of what was made, when, from what, by whom.

That is the honest one-paragraph version. The longer version is that "MES" covers an enormous range, from a module that tracks work orders to a plant-wide platform that took three years to deploy. This post explains what the category actually does, where it sits, and when you need it.

What does an MES actually do?

The core functions of an MES are consistent across vendors, and roughly track the manufacturing operations management activities described in the ISA-95 standard:

Where does MES sit? The ISA-95 view

ISA-95 (adopted internationally as IEC 62264) is the standard for integrating enterprise and control systems, built on a five-level hierarchy often drawn as the Purdue model. The standards body ISA describes the family at isa.org: Level 0 is the physical process, Levels 1-2 are sensing and control (PLCs, SCADA/HMI), Level 3 is manufacturing operations management, where MES lives, and Level 4 is business planning, where ERP lives. Most of what ISA-95 standardizes is the conversation between Levels 3 and 4: what the ERP tells the MES to make, and what the MES reports back.

The plant software stack (ISA-95 levels)LEVEL 4 · ERPOrders, inventory, money · horizon: weeks-monthsLEVEL 3 · MES / MOMExecute + track production · horizon: shifts-daysLEVELS 1-2 · SCADA / HMI / PLCsSense + control machines · horizon: secondsLEVEL 0 · PHYSICAL PROCESSMachines, material, people, paperAI OPERATIONALLAYERConnects dataacross everylevel + paper +tribal knowledge.Search, cite,act. Norip-and-replace.ISA-95 levels on the left; an AI layer reads and acts across all of them.
The plant software stack. MES occupies ISA-95 Level 3; an AI operational layer works across levels rather than replacing any of them.

MES vs ERP vs SCADA vs AI operational layer

ERPMESSCADAAI operational layer
Core questionWhat should we make, buy, and bill?What is being made right now, and how?What is this machine doing this second?What does all of it mean, and what should happen next?
Time horizonWeeks to monthsShifts to daysSecondsSeconds to months
Primary usersFinance, planning, purchasingProduction supervisors, qualityOperators, controls engineersEvery role, in role-specific views
Data it ownsOrders, inventory, costsWork orders, genealogy, OEEProcess values, alarmsNone exclusively; it connects ERP, MES, machines, paperwork, and tribal knowledge
Typical failure modeDescribes the plant as planned, not as runLong, rigid implementationsData trapped at the machineOnly as good as the sources it connects
The layers answer different questions; none replaces the others.

The comparison with a CMMS is similar: the CMMS owns maintenance work, the MES owns production execution, and they should talk.

How do you evaluate an MES? Seven questions

  1. What problem, in dollars, are we solving? Traceability mandates, scrap, schedule chaos, and audit prep point to different scopes. "Visibility" alone is not a scope.
  2. What must it integrate with, and how? Get specific about your ERP version, your PLC brands, and who owns the interfaces when something changes.
  3. Who maintains it after go-live? An MES that needs a vendor engineer for every routing change will freeze your process in place.
  4. What is the operator experience? If data entry at the station is slower than the paper it replaces, operators will defeat it, and your data will be fiction.
  5. Can we deploy one line first? Insist on a scoped pilot with defined success metrics before a plant-wide commitment.
  6. What is the true total cost? License, hardware, integration, internal labor, and the multi-year tail of change requests. Get references from plants your size.
  7. What happens to the data? You should be able to get your own production history out, in bulk, in an open format. Anything less is a hostage situation.

When do you need an MES, and when is it overkill?

You likely need MES-grade capability when regulatory traceability is mandatory (pharma, medical devices, aerospace, some food), when you run high-volume discrete production where minutes of misrouting are expensive, or when a corporate customer contractually requires electronic lot genealogy.

For many mid-market plants, a full MES is overkill, at least as a first step. The classic failure pattern: an 18-month implementation scoped around last year's process, a rigid data model, and a floor that quietly returns to whiteboards. If your actual pain is that production data lives on paper and in heads, that OEE is estimated rather than measured, or that nobody can answer "what happened on line 2 last night" without three phone calls, the fix is a data foundation, not a monolith. That is the wedge Harmony takes as a manufacturing operating system: digitize the paper first, connect the ERP, MES, or QMS you already have, pull machine signals, and compute true OEE from the source, with no rip-and-replace. Plants like Chattanooga Labeling Systems replaced paper production logging with real-time visibility without a classic MES project.

An honest rule of thumb: buy an MES when a specific, named requirement demands Level 3 execution control. Build the data layer first when your problem is knowing what is happening at all.