A Manufacturing Execution System (MES) is the software that runs a factory floor in real time. It sits between the Enterprise Resource Planning (ERP) that plans what to build and the machines that actually build it, tracking work-in-process, enforcing quality and compliance, capturing production data, and giving operators and managers a live picture of what's happening on the line.
This article provides a comprehensive overview of MES, the software layer essential for modern Industry 4.0 operations. Learn where it sits in the manufacturing stack, why it matters, and where its blind spots are - including the gap that vision AI is now filling on real production lines. For any manufacturer looking to boost efficiency and maintain a competitive edge, understanding and implementing an MES is a vital step toward a smart, data-driven future.
What Is a Manufacturing Execution System?
A Manufacturing Execution System (MES) is a specialized software layer that controls, monitors, and records physical production processes on the factory floor in real-time.
Often described as the brain or central nervous system of manufacturing operations, an MES bridges the gap between high-level business planning systems, like Enterprise Resource Planning (ERP), and the actual automation and machinery on the shop floor. MES answers the question: "What is happening on the floor right now, and is it going to plan?"
According to the ISA-95 global standard, an MES operates at Level 3, translating high-level production plans into specific, real-time execution tasks to ensure production stays on track from start to finish.
- Level 0-2: the physical process, sensors, and controllers (PLCs, SCADA)
- Level 3: MES, the manufacturing operations layer
- Level 4: business planning and logistics (ERP)
Where MES Sits in the Manufacturing Stack
A typical large manufacturer runs a layered stack:
- ERP: orders, finance, inventory at the business level, materials planning
- MES: production execution, quality, traceability
- SCADA / Historian: supervisory control and time-series data capture
- PLC / DCS: programmable controllers running individual machines and cells
- Sensors, actuators, robots, vision systems: the physical layer
Information flows up and down. A work order leaves ERP, lands in MES, gets dispatched as a job to a line, runs through PLCs and SCADA, and the resulting production data flows back up to MES and on to ERP.
Challenges Solved by an MES
Before the widespread adoption of MES, many manufacturers operated with significant blind spots. An MES addresses several critical operational hurdles:
- Paper-Based Processes: Traditional manual tracking is slow, inconsistent, and highly prone to human error.
- The Black Box Effect: Without an MES, the plant floor is often a "black box" to executives, where data is only reported in delayed, summarized chunks rather than real-time events.
- Data Silos: Many facilities use disparate software for inventory, quality, and maintenance that do not communicate, leading to massive inefficiencies.
- Information Latency: Delays in reporting mean that issues like machine breakdowns or quality defects might not be identified until hours or days after they occur, causing expensive waste.
An In-Depth Look at MES Functions
Across MESA-11, the functions plants use most include:
- Production scheduling and dispatch. Turning the ERP-level production plan into a sequence of jobs across lines, cells, and operators. Reassigning when a machine goes down, materials are late, or priorities change.
- Work-in-process (WIP) tracking. Knowing which units are at which station, in which state, attached to which work order. This is the backbone for every other function.
- Quality management. Pulling SPC data, logging inspections, capturing non-conformances, holding lots, triggering corrective actions. Connected to the quality team's CAPA process.
- Genealogy and traceability. Recording the materials, machines, operators, and process parameters tied to every serialized unit. Essential in regulated industries (food, pharma, aerospace, automotive, defense) and increasingly required by customers in others.
- Performance and OEE. Computing Overall Equipment Effectiveness and surfacing the losses. This is usually the number leadership watches.
- Maintenance and downtime. Logging unplanned stops, root causes, and time to repair. Often coordinated with a separate CMMS or EAM.
- Labor and electronic work instructions. Showing operators what to do at each step, capturing sign-offs, enforcing skills and certifications.
- Compliance and electronic records. 21 CFR Part 11, IATF 16949, AS9100, HACCP - depending on the industry. MES is where the audit trail lives.
A plant might do all of them well, or use MES for a few and rely on point tools for the rest. The variation across sites is part of what makes large MES rollouts hard.
Computer Vision: The Eyes of the MES
Modern computer vision is further changing the math. Cameras are cheap, edge compute is cheap, and the models that interpret images have crossed the threshold where a plant team can train them on the parts and conditions that matter on their line.
A vision AI system sits alongside MES, not inside it. The camera and model do the perception. The MES does what it has always done: record the result, hold the lot, dispatch the rework, update the genealogy. The integration points are familiar: REST APIs, MQTT, OPC UA, direct database writes, depending on the MES.
The use cases that come up most often in conversations with manufacturing teams include:
- Defect detection on the line. A camera and model inspect every unit at speed - surface defects, missing components, dimensional issues, color and print verification. Results write back to MES as pass/fail with images attached for genealogy. USG uses Roboflow for defect detection and predictive maintenance on building-materials lines.
- PPE and safety compliance. Cameras at zone entries flag missing hard hats, safety glasses, high-vis, or unauthorized presence in a restricted area. Events log into MES or directly into the safety system.
- Inventory and asset tracking. Counting and identifying assets - railcars, pallets, finished goods, raw stock - without manual scans. BNSF, North America's largest freight rail operator, uses Roboflow to automate yard inventory across rail facilities.
- Process and step verification. Confirming a work-instruction step was completed correctly - the bolt is torqued, the label is applied, the component is in the right orientation - before the operator can advance the job in MES.
- Cycle-time and bottleneck analysis. Counting throughput at each station against the standard, feeding live performance into the MES's OEE calculation.
In each case, vision AI is doing the perception layer that MES needs but was never designed to do itself. MES remains the system of record.
What It Takes to Ship
Manufacturing teams have heard the vision AI pitch before. Most have a failed pilot somewhere in their history: a custom dev shop that built something brittle, a research team that never got past one line, an enterprise platform that demoed well and stalled in procurement.
Three things separate the deployments that ship from the ones that don't.
The domain expert is in the loop. The plant quality engineer knows what a defect looks like better than any AI researcher. The platform has to let them label data, train a model, and iterate without a PhD in computer vision. The work is concrete: inspect this kind of weld, count these kinds of parts, and the people who know what good looks like need to be the ones building the system.
Cloud, edge, and on-prem from the same platform. Plants are not cloud-only environments. Lines need low-latency inference at the camera. IT and OT have rules about what data leaves the facility. The platform has to deploy to the cloud, to a Jetson or industrial PC at the edge, or to an on-prem server - without rebuilding the pipeline each time.
The platform integrates with what's already running. MES, SCADA, PLCs, historian, ERP. Vision AI that can't write back to the systems of record is a science project. Vision AI that writes a defect event into MES and triggers the existing CAPA workflow is a production system.
This is the work Roboflow does with enterprise manufacturers. Over half of the Fortune 100 build with Roboflow today, and more than 55 billion model inferences run through the platform across critical industries.
If you're scoping a vision AI deployment alongside an existing MES, request a Roboflow demo today.
Common Questions About MES
1. What is the difference between ERP and MES?
ERP is like the general manager focused on business planning (finance, sales); MES is the head coach focused on real-time execution (machinery, operators, quality).
2. Can an MES integrate with older equipment?
Yes. Modern systems connect to existing PLCs (Programmable Logic Controllers) and sensors via standard protocols like MQTT or OPC-UA to bridge legacy machines with digital dashboards.
3. Is an MES too complex for small manufacturers?
No. Modular, cloud-native platforms now allow smaller companies to start with individual modules (like quality or OEE) and scale to a full MES as they grow.
Cite this Post
Use the following entry to cite this post in your research:
Contributing Writer. (Mar 1, 2026). What an MES Does on the Factory Floor. Roboflow Blog: https://blog.roboflow.com/manufacturing-execution-system/