How to Integrate ERP, MES, and Shop-Floor Data

This is the core of data engineering, and the leap from a disconnected floor to a connected one.

Why it's hard: the OT/IT divide

Integration is harder in manufacturing than in most industries because you're bridging two different worlds. Operational technology (OT) — your PLCs, SCADA, and MES — speaks industrial protocols, runs in real time, and often uses proprietary or vendor-specific formats. Information technology (IT) — your ERP and QMS — speaks databases and APIs, and often works in batches. The two rarely share common keys, common units, or a common clock.

That gap is the leading technical reason manufacturing AI stalls: RAND's analysis put manufacturing AI failure around 76%, with OT/IT integration and data quality named as the top causes (RAND, 2025). Bridging OT and IT cleanly is exactly what separates a foundation that works from a pile of disconnected systems.

What each system holds

Integration is really reconciliation, so it helps to know what you're reconciling:

• ERP — orders, inventory, scheduling, cost: what was planned.
• MES — work orders, sequence, output, rate: what's actually being produced.
• SCADA / PLCs — temperatures, pressures, cycles, faults: the real-time process.
• QMS — inspections, scrap, rework: quality.

Each holds one piece of the truth. (For a deeper breakdown of the gaps between them, see ERP, MES, SCADA: what data each system holds.)

How to integrate it, step by step

Step 1 — Map what you have

Inventory every source: which systems, what data they hold, what format and protocol they use, and where the gaps are. You can't connect what you haven't mapped. This is the audit at the heart of data readiness.

Step 2 — Define a common model

Decide, once, what the shared entities and definitions are: a part number is the same part everywhere, a "shift" has one meaning, a "good unit" is counted the same way on the floor and in the back office. This common model is the backbone of reconciliation — without it, you're connecting systems that still disagree.

Step 3 — Build the pipelines

Move data out of each system automatically. On the OT side, a standard like OPC-UA bridges PLCs and SCADA into a common stream; on the IT side, APIs and connectors pull from ERP and MES. Use streaming for real-time floor data and batch (ETL/ELT) where it fits. These data pipelines replace the brittle weekly export. (Primer: Data pipelines for manufacturing.)

Step 4 — Land it in one foundation

Pipe everything into a single governed data warehouse or lakehouse, where it's cleaned, reconciled to the common model, and structured for use. This is the single source of truth the whole operation reads from.

Step 5 — Govern it

Apply data governance: one agreed definition per metric, role-based access, and an audit trail. This is what keeps the integrated foundation trustworthy — and it's non-negotiable in regulated sectors.

Step 6 — Keep it live and healthy

Integration isn't one-and-done. Pipelines need monitoring, new sources get added, and definitions evolve. Tending the foundation is the work of continuous optimization — neglect it and it decays back toward dark data.

Common pitfalls

The mistakes that sink integration projects:

• Trying to rip and replace. You don't need to scrap a working ERP or MES. Good integration builds around existing investments.
• Dumping data into a lake with no common model. Connection without reconciliation just creates a bigger, faster mess — not a foundation.
• Relying on manual exports. A spreadsheet pulled by hand is stale the moment it's saved. Integration has to be automated to stay live.
• Skipping governance. Without one definition and clear access, the "integrated" data drifts back into conflicting numbers.
• Underestimating OT. Floor protocols and proprietary formats are genuinely harder than business-system APIs. Plan for them.

The payoff

Integration is the leap from Disconnected to Connected — Stage 1 to Stage 2 on the Data Maturity Model. It's the most important jump in the model, because it's what finally makes one number trustworthy and gives BI and AI something solid to run on. Poor data quality alone costs organizations an average of $12.9 million a year (Gartner); integration is how you stop paying that tax.