The trap
Every MES project starts with the same promise: real-time visibility into what is happening on the shop floor. Every line, every machine, every shift. OEE trending upward. Downtime root causes identified within minutes, not weeks.
Most of them fail to deliver it — not because MES software is bad, but because the data problem is harder than it looks from the project kick-off.
The trap is this: shop-floor data is fragmented, inconsistent, and frequently wrong. An MES that ingests fragmented data doesn't generate visibility. It generates a more elaborate display of fragmentation.
The projects that succeed are the ones that treat data quality as the primary deliverable and MES configuration as the secondary one.
What fragmented shop-floor data actually looks like
A typical mid-size manufacturer has:
- PLC/SCADA data — machine signals at millisecond granularity, often in vendor-proprietary formats, with no consistent naming convention between machine types or vendors
- Paper records — shift logs, downtime reasons, quality checks, filled out after the fact and keyed into spreadsheets weekly
- ERP data — production orders, BOM, routing — often delayed by a day or more as production records are confirmed
- Maintenance records — CMMS systems tracking work orders, but not systematically linked to production interruption events
These systems don't share identifiers. An OEE calculation requires linking a PLC downtime signal (machine ID: PRS-L2-03) to a production order (in ERP: WO-20241105-0042) to a downtime reason (paper record: "wait for material" written at end of shift). None of those three identifiers refer to the same thing in the same language.
The four-step framework
Step 1: Data inventory before tool selection
Before evaluating MES vendors, map what data exists, where it lives, what format it's in, and how reliable it is. This is slow, unglamorous work. It almost always reveals that the "real-time machine data" the project assumed exists is actually batch-exported CSV files from the SCADA system once a day.
The output of this step is a data inventory: for each data element required for the business use case (typically OEE, production counts, downtime reasons, quality defects), document the source system, latency, format, and known quality issues.
Step 2: Define the minimum viable data pipeline
Most MES projects scope for full integration on day one. That is how you end up 18 months in with a go-live still 6 months away.
Define the minimum data pipeline that delivers the highest-priority use case. Usually this is OEE on the two or three highest-volume production lines. Get that working and producing reliable numbers before expanding.
A minimum viable pipeline for OEE might be: PLC signal → tag historian → OEE calculation engine → dashboard. No ERP integration. No automated downtime reason capture (manual entry to start). No quality integration.
That pipeline can be live in 8-12 weeks. The ERP integration and automated downtime capture come in later phases, against real operational data that shows where the gaps actually are.
Step 3: Operator adoption before data completeness
An MES that operations teams don't trust is worthless. The fastest way to lose operator trust is to show OEE numbers that don't match what operators know to be true.
Before going live on any line, validate the OEE calculation with the supervisors who run that line. Run the calculation against historical data and ask them to rate the accuracy. Fix the discrepancies — usually tagging issues or edge cases in state machine logic — before showing numbers to management.
Operators who were involved in validation become advocates. Operators who see numbers they don't trust become resistors.
Step 4: Enrichment in layers
Once the minimum pipeline is live and trusted, add enrichment in layers. Each layer adds a new data source and makes the existing data more useful.
Layer 1 (base): PLC signals → OEE. Real-time availability, performance, quality on each line.
Layer 2: ERP integration → OEE with production order context. Now you can slice OEE by product, customer, and work order.
Layer 3: Downtime reason capture → pareto of downtime causes. Either manual (operator entry on touchscreen) or automated (maintenance event lookup from CMMS).
Layer 4: Predictive layer → leading indicators. Time-series analysis on machine signals to identify patterns that precede unplanned downtime.
Each layer is a separate project scope with its own ROI calculation. Projects scoped this way ship. Projects scoped to do all four layers simultaneously rarely do.
The vendor question
Most MES vendors will tell you their platform handles the data integration challenge. Some of them are right, if your machines are modern, your PLC vendors are on their supported list, and you're willing to pay for the implementation services.
Many manufacturers find that their machine estate is older than the vendor's sweet spot, their PLC vendors are not on the preferred list, and the implementation services quote is larger than the software license.
In those cases, the practical approach is often a thinner integration layer (an OPC-UA server or an IIoT gateway) that normalizes PLC data before it reaches the MES, and a MES configured against normalized data rather than raw machine signals.
We've implemented this pattern across a range of machine estates, from CNC machining centers with modern Fanuc controls to 1990s-era injection moulding machines with no native connectivity. The approach is the same; the implementation details differ.
What success looks like
A successful MES implementation, measured 12 months post go-live:
- OEE numbers that supervisors and managers agree are accurate
- Downtime pareto that operations teams act on (changes to PM schedules, changeover procedures, or staffing)
- Production reporting that is faster than the previous process, not slower
- An integration team that knows how to add the next line themselves, without a consultant
The last point matters. The goal is a capability you own, not a system you depend on us to run.
This article reflects Vatsa Solutions' manufacturing practice methodology, developed across MES and OEE implementations in automotive, plastics, and food & beverage manufacturing environments.