Most manufacturing plants do not suffer from a lack of software. They suffer from a gap between two pieces of software that were never designed to talk to each other. The ERP knows what should be produced this month. The PLCs know what the machines did in the last second. Between the two sits a layer that, in a surprising number of factories, is still made of spreadsheets, printed work orders and a whiteboard.
That layer has a name — Manufacturing Execution System — and a formal place in the ISA-95 model, where it occupies Level 3, between business planning above and process control below. What makes it interesting is not the acronym. It is that almost every recurring shop-floor problem turns out to live in that gap.
What a MES actually owns
The clearest way to understand the boundary is by the question each layer answers and the time horizon it answers it over.
| Layer | Question it answers | Time horizon |
|---|---|---|
| ERP (Level 4) | What should we make, for whom, with which materials? | Weeks to months |
| MES / MOM (Level 3) | What is being made right now, by whom, from which lot, at what quality? | Minutes to shifts |
| SCADA (Level 2) | What are the machines doing and are they within setpoints? | Seconds |
| PLC / sensors (Levels 0–1) | Open the valve, read the temperature, stop the conveyor | Milliseconds |
Read that table and the failure mode becomes obvious. When Level 3 is a spreadsheet, nothing connects the production order to what physically happened. The ERP eventually receives a quantity, hours after the fact, typed by a supervisor. Everything in between — the lot genealogy, the downtime, the deviations, the rework — exists only as paper, if it exists at all.
Five signals that a plant has outgrown its spreadsheets
- Month-end takes three days. Not because the accounting is complex, but because production figures have to be reconciled by hand against what the ERP thinks was consumed.
- Nobody can answer a traceability question in under an hour. The data exists across four systems and one filing cabinet, and someone has to physically assemble it.
- OEE is a monthly number, not a live one. By the time a performance drop is visible in a report, the batch that caused it has shipped.
- Rework is invisible. It happens, it costs labour and material, and it appears nowhere in the cost of the order.
- Every audit triggers a project. If preparing for a customer audit means two weeks of document assembly, the records are being reconstructed rather than maintained.
None of these is a software problem in isolation. Together, they describe an execution layer that is being run by human memory.
What changes in the first ninety days
Deployments that go well tend to share a shape: one line, one product family, one clearly stated problem — not a site-wide rollout. Within a quarter, the visible changes are consistent across industries:
- Declaration disappears. Machine states are captured automatically and timestamped; operators qualify exceptions instead of logging everything, which usually gives them time back rather than costing them time.
- Lot genealogy becomes a query. Raw material lots, machine parameters, operator, and finished serial numbers are chained at the moment of production, so a recall becomes a search rather than an investigation.
- Deviations get caught in-process. A parameter drifting out of tolerance raises an alert on the line, not a non-conformity three days later in the lab.
- The electronic batch record assembles itself. It is built continuously as the batch runs, which is the only version an auditor genuinely values, because it carries an audit trail instead of a signature added afterwards.
This is the scope covered by platform vendors positioned at Level 3 — UXP Corp MES software being one of the European examples, built as separate usage modules rather than as a monolith, which is what makes a single-line pilot feasible in the first place.
Regulation is quietly making the decision
For a long time, the business case rested on productivity alone, which made it easy to postpone. That is changing, because the compliance deadlines are now dated.
- In the EU, pharmaceutical serialisation has been mandatory since February 2019 under Delegated Regulation (EU) 2016/161: a unique serial number on every pack, verified at dispensing.
- In the US, FSMA Section 204 requires detailed traceability records — key data elements at critical tracking events — for foods on the FDA's Food Traceability List. The FDA extended the compliance date to 20 July 2028, which sounds distant until you count back the time needed to instrument a plant.
- For medical devices, EU 2017/745 ties unique device identification to registration in EUDAMED.
What these texts have in common is granularity. They do not ask whether a plant has records; they ask for records at the level of the event, the lot and sometimes the individual unit, retrievable on demand. That requirement is structurally incompatible with an execution layer made of spreadsheets — which is why traceability, rather than productivity, is now the entry point for most projects. Vendors such as UXP Corp have reorganised their offering around exactly that shift.
The practical first step costs nothing: take one production order, and try to reconstruct from your current systems which raw material lots went into it, which machine settings applied, who signed off, and how long the line was stopped. The time that exercise takes is the size of your gap.