Ask an ERP what's happening on the floor right now and it will tell you what was scheduled. Ask an MES and it will tell you what's actually running, on which machine, built by which operator, against which lot. That gap between "planned" and "actual" is the entire distinction, and it's why plants that try to run execution out of their ERP eventually stop trying.
The comparison, side by side
| ERP | MES | |
|---|---|---|
| Time horizon | Days to months | Seconds to minutes |
| Transaction unit | Purchase order, sales order, work order as a whole | Individual unit, operation, scan, machine state |
| Data latency | Batch-posted, often end of shift or overnight | Real time |
| What it answers | What should we build, what did it cost, what do we owe | What's running right now, is this unit in spec, where did this lot go |
| Genealogy | Lot-level at best, usually none | Unit-level: machine, operator, timestamp, parameters |
| System of record for | Finances, planning, inventory valuation | Execution, quality, equipment state |
Neither system is the lesser one. ERP is right that a plant doesn't need second-by-second precision to decide what to build next month. MES is right that a plant does need it to decide whether the unit on the line right now should ship.
What breaks when a plant runs execution out of ERP
Three things, consistently. Data arrives batch-posted instead of live, so a supervisor checking the ERP at 10am is looking at what happened on the prior shift, not what's happening now. Genealogy stops at the lot: an ERP can tell you which purchase order a batch of steel came from, not which machine, operator and process parameters touched the specific unit that just failed a customer inspection. And there's no real-time state, so questions like "is line 4 actually running" get answered by walking the floor, not by querying the system meant to know.
The cost of that gap is visible in what MES-equipped plants can do that ERP-only plants can't. BIC's Tunisian facility runs over 100,000 finished-good transactions in a single shift on Plex MES. An ERP's data model is built to post a work order once, as a whole, when it closes, not to absorb a discrete transaction for every one of 100,000 units as each one completes. Try to run that volume through an ERP's order-based schema and you either drown the system or you stop recording at the unit level and go back to end-of-shift estimates, which is exactly the visibility gap MES exists to close.
Genealogy shows the same pattern from the quality side. Before it had unit-level traceability, an anonymous tool manufacturer needed 5 days to investigate a defect; an ERP's lot-level records were the ceiling on how precise that investigation could get. With MES-level genealogy, the same investigation took 30 minutes and avoided $266,000 in rework, because the system already knew which machine and operator touched the failing unit instead of forcing someone to reconstruct it.
When a plant genuinely doesn't need one
Low mix, single line, one shift, and nobody has ever needed to trace a unit past its lot: an ERP plus a whiteboard is a legitimate operating model, and adding MES on top of it buys enforcement for a process that was never inconsistent enough to need enforcing. The signal to watch for is disagreement, not scale. The moment two shifts run the same job differently, or a customer asks a traceability question the ERP can't answer, the gap MES fills stops being theoretical.
The integration boundary is where the money goes
Most plants end up running both, and the interface between them, work order release from ERP to MES, completions and consumption flowing back, inventory sync, quality holds, is where implementation budgets actually go. Two paths avoid rebuilding that interface as custom middleware. One is buying ERP and MES from the same vendor stack, which is what Precision Coating did with Plex ERP, MES, and QMS and what Bayrak Technik did on the same stack in automotive: the interface is pre-built and maintained by one vendor instead of negotiated between two. The other is accepting a genuinely custom integration, which costs more up front but doesn't lock the plant into one vendor's roadmap for both systems. Neither path is free. The plants that get surprised are the ones who scoped the MES purchase and treated the ERP interface as a detail to figure out later.
Where to go next
- For what MES does on its own, independent of the ERP question, see What Is MES?
- The process optimization use cases and benchmarks cover outcomes across the 416-deployment corpus these systems feed into.
- Vendors that ship both layers, or integrate cleanly with what's already on the floor, are ranked on the process optimization software hub.
- Newman Technology hit 98% inventory accuracy after its MES started reconciling against the ERP automatically instead of by manual count.
AI Roadmap — Eight workflows in your plant where AI would pay back this quarter. We know which. 12 pages, in 72 hours.