SCADA supervises machines in real time (seconds), MES executes and tracks production (minutes to hours), and ERP plans and records the business (days to weeks). They are three stacked layers on three different timescales, and the gaps between them, where data has to be handed off, are exactly where plant data silos form.
These three acronyms get used interchangeably by people who should know better, and the confusion is expensive: buy the wrong layer for a problem and you will spend years asking one system to do another's job. This is a plain comparison of what each actually does, how they stack, where the hand-offs leak, and how to decide which layer owns a given job. No products are being sold here, every plant of a certain size runs some version of all three.
What does SCADA do?
SCADA supervises the physical process in real time. Standing for Supervisory Control and Data Acquisition, it collects live readings from PLCs and instruments, shows operators the state of the process on screens, raises alarms, logs history, and sends supervisory commands, a setpoint change, a start, a stop, back down to the equipment. Its timescale is seconds. It cares about the process: is the temperature right, is the pump running, did an interlock trip.
What SCADA does not know is the business context around the process. It can tell you the filler stopped and which alarm fired; it cannot tell you which order was running, what the stop cost, or whether the same failure happened last month. That is by design, SCADA's job is supervision, and it does it well. For the full picture, see what SCADA is.
What does MES do?
MES manages production execution, the layer between the machines and the business. A Manufacturing Execution System tracks what is being made against what was ordered: it dispatches work, records what actually ran, captures quality results, and maintains genealogy (which materials and steps went into each unit). Its timescale is minutes to hours. It cares about the order and the product: what should be made, what was made, and whether it was good.
MES sits above SCADA and below ERP, translating between them. It takes the business's plan and turns it into floor instructions, then takes the floor's reality and turns it into production records the business can use. When people say a plant has a "gap between the office and the floor," they usually mean the MES layer is thin or missing. See what MES is for the detail.
What does ERP do?
ERP runs the business, not the floor. Enterprise Resource Planning is the company's integrated system of record: sales orders, inventory, purchasing, finance, and high-level production planning, all in one database. Its timescale is days to weeks. It answers what was promised, what it costs, and what is in stock. When finance closes the month or planning explodes a bill of materials into purchase orders, that is ERP.
ERP is built for transactional accuracy and planning horizons, not second-by-second execution. Ask it to run the floor and you get operators keying production data in hours late, which makes the ERP's picture of the plant both a burden to maintain and wrong. The classic mistake is expecting the system designed for financial truth to deliver real-time responsiveness it was never architected for. More in ERP for manufacturing.
How do SCADA, MES, and ERP compare?
The cleanest way to see it is side by side, because the differences are really about timescale and what each layer cares about.
| SCADA | MES | ERP | |
|---|---|---|---|
| Job | Supervise the process | Execute & track production | Plan & record the business |
| Timescale | Seconds | Minutes–hours | Days–weeks |
| Cares about | The process | The order & the product | The transaction |
| Purdue level | Level 2 | Level 3 | Level 4 |
| Key question | "Is it running right?" | "What was made, and was it good?" | "What did we promise and what does it cost?" |
| Data | Live tags, alarms, history | Work orders, genealogy, quality | Orders, inventory, finance |
Where are the hand-offs, and why do they leak?
The gaps are at every seam between layers, and they leak because the systems were bought separately and never designed to hand data to each other in real time. ERP was bought by finance, MES by operations, SCADA by controls engineering. Each is authoritative for its own slice, and each speaks its own data model. The most valuable questions, why margin was low on line 3 last Tuesday, live precisely in the seams, where no single system owns the answer.
So people become the integration. An operator reads a number off the SCADA screen and types it into the MES. A supervisor exports an MES report and re-keys totals into the ERP. This "swivel-chair integration" is slow, error-prone, and the clearest symptom of the problem covered in manufacturing data silos. The systems are all correct; the plant still cannot answer a simple cross-layer question without a human stitching screens together.
How do you decide which layer owns a job?
You decide by matching the job's timescale and subject to the layer built for it. Ambiguity here is what creates duplicate data entry that lasts forever, so it is worth settling explicitly.
- Name the timescale. Seconds means SCADA, minutes-to-hours means MES, days-to-weeks means ERP. Speed is the fastest sorter.
- Name the subject. The process is SCADA, the order and product are MES, the transaction and money are ERP.
- Find the current owner. Ask which system holds the authoritative version today. If two claim it, you have found a silo.
- Assign one system of record per fact. Every number should have exactly one home; the others reference it rather than re-entering it.
- Kill the re-keying at that seam. Wherever a person is retyping to bridge two layers, that hand-off needs a real connection, not a swivel chair.
- Check it against the map. Confirm your assignment matches the layered structure of the Purdue model which puts SCADA at level 2, MES at level 3, and ERP at level 4.
What closes the gaps between them?
A real-time operational layer that connects the machines and all three systems into one live shared model. SCADA, MES, and ERP are each doing their own job well; the failure is between them, so the fix is not a fourth system that replaces them but a connective layer that reads from each and joins their data. It captures the floor's reality from SCADA and the machines, ties it to the order in MES and the cost in ERP, and makes the whole picture answerable in one place, the idea behind a manufacturing operating system.
This is where the timescale story pays off. When the connective layer captures data in real time rather than nightly, the floor's reality flows into MES and ERP accurately and on time instead of being re-keyed late. The cross-layer question stops requiring four logins and a spreadsheet. And because the layer reads rather than replaces, no existing system has to be ripped out to get there, the SCADA keeps supervising, the MES keeps executing, the ERP keeps its books, and the connective layer simply makes them agree.
By the numbers
The three-layer structure is not a vendor invention; it is codified. The ISA-95 / IEC 62264 standard for enterprise-control integration defines exactly these levels and the interfaces between them, and the NIST Systems Integration Division studies the very hand-offs that leak. The recurring finding across manufacturing-technology research is that the shop floor and the business systems are chronically out of sync, a gap rooted in the timescale mismatch, not in any one system being bad at its job.
Where Harmony fits: Harmony does not replace your SCADA, MES, or ERP. It connects to them, and to your machines and paperwork, as one real-time operational layer, computing true OEE from source signals and joining process, production, and business data so the questions that used to require four logins get answered in one place, with the control layer untouched. No rip-and-replace (see the connected systems module), and a real example lives in how CLS unified its floor.
Where does this fit with the rest of the stack?
SCADA, MES, and ERP are the middle of a larger picture. Below them are the sensors and networks like Profinet and Profibus that carry raw signals up; around them is the IIoT movement pushing more data off the floor faster; and the whole arrangement is mapped by the Purdue model. Understanding which layer owns which job is the difference between a plant where the systems cooperate and one where people spend their days being the glue. For how production plans turn into floor schedules across these layers, see production scheduling.