How Siemens built a unified manufacturing data backbone by integrating MES and ERP

Siemens' Amberg plant in Bavaria manufactures programmable logic controllers, the SIMATIC line, and has done so since 1989. By the early 2010s the plant was processing roughly 15 million components daily across a highly automated floor, with machine utilisation rates that most manufacturers treat as aspirational targets. The data problem was not a shortage of signals: SCADA systems, quality inspection stations, and warehouse management tools were all generating readings. The problem was that none of those readings spoke the same language as SAP, which ran financials, procurement, and production orders. A defect caught at a test station stayed inside the MES until someone typed it into a form. A change to a bill of materials in SAP did not propagate to the floor until a shift supervisor printed a new routing sheet. The seam between the two systems was staffed by humans doing copy-paste work, and every hour of latency in that seam was an hour of potential waste that no dashboard could see.

What they did

Siemens invested in bidirectional, event-driven integration between its MES layer (running on its own Opcenter software, then called SIMATIC IT) and SAP ERP. The architecture rested on two decisions that are easy to understate.

First, they treated master data as a single source of truth from the ERP downward. Materials, equipment hierarchies, and bill-of-materials structures were maintained exclusively in SAP and published to the MES via a defined integration layer. This sounds obvious, but in practice most plants maintain shadow copies of BOMs in the MES because the ERP version is either too coarse or too slow to update. Siemens chose to fix the ERP data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → rather than work around it, which required months of master data remediation before any integration code was useful. If you want to understand why that remediation step is so often underestimated, thework involved in aligning materials, BOMs, and equipment hierarchies across a multi-line plant is the right place to look.

Second, they inverted the production order flow. Rather than the MES generating its own job queue and reconciling with SAP after the fact, production orders originate in SAP and the MES receives them in near real time. Actual quantities, scrap codes, and labor confirmations flow back to SAP automatically at each operation step, not at shift end. The result is that the financial ledger and the physical floor are, within minutes, describing the same reality.

The Amberg implementation also connected quality data to traceability records in a way that satisfied both internal process needs and external regulatory obligations. Each unit leaving the line carries a serial number linked to every component serial and every process parameter recorded during assembly. When a product liability question arises, or when a customer in a regulated industry needs a conformance record, that trace is available in minutes rather than days.

On the infrastructure side, Siemens moved away from flat-file batch transfers and point-to-point database links toward a middleware layer (in their case built on their own integration platform) that exposed standardized APIs. This mattered because the plant floor is not static: new machines arrive, new product variants require new routing steps, and any architecture that requires custom code for every change becomes a maintenance liability within two or three equipment refresh cycles.

The results

Siemens has published figures for Amberg that are specific enough to be useful. The plant achieved a product quality rate of approximately 99.9988 percent, meaning fewer than 12 defects per million units. Machine utilisation runs above 75 percent across the automated lines, which is exceptional for a plant producing over 1,000 product variants. The plant also doubled output between 1990 and the mid-2010s while holding headcount roughly flat, though it is worth being clear that the MES-ERP integration was one contributor among several, including broader automation investment.

What the integration specifically produced in data terms: the gap between a defect occurring on the floor and that defect appearing in a reportable production record dropped from hours to under five minutes. Engineering change orders, which previously took two to three days to propagate from SAP into active work instructions on the floor, now complete in under an hour. Neither figure has been independently audited for public consumption, so treat them as directionally accurate rather than certified benchmarks.

The traceability architecture also reduced recall exposure in a measurable way. When a component supplier issued a quality advisory in 2018, Amberg could identify affected serial numbers and their customer destinations within a single shift. A plant running batch reconciliation between MES and ERP would typically need three to five days for the same exercise, during which finished goods sit under a hold and customer commitments slip.

What transfers

The Amberg case carries three lessons that do not depend on being Siemens.

Master data remediation is not a pre-project task, it is the project. The integration layer is worthless if materials in SAP are coded differently from materials in the MES. At Amberg, the remediation of equipment hierarchies and BOM structures took longer than the integration development itself. CDOs who try to shortcut this step end up with a pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition → that moves dirty data faster.

The direction of data authority matters as much as the connection itself. Siemens chose SAP as the system of record for everything that had a financial consequence, and the MES as the system of record for process parameters and quality events. That boundary is not arbitrary: it maps to who owns the data governancedata governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition → accountability. Finance owns the BOM because they own the standard cost. Quality owns the process record because they own the conformance certificate. When you let both systems maintain their own versions of the same object, you get reconciliation work, not integration.

Thegap between what your systems measure and what actually drives OEE and scrap rates is usually where the business case lives. At Amberg, the integration justified itself on reduced expediting costs and faster engineering change cycles, not on the analytics layer built afterward.

Where your context differs: Amberg is a greenfield-friendly, high-volume, low-mix-by-Siemens-standards plant. If you run a job shop with hundreds of active work orders and frequent engineering changes, the event volume and the change management overhead of this architecture scale differently. The principles hold; the implementation timeline and the middleware sizing do not.

The one thing Siemens could do that many manufacturers cannot is deploy their own MES product, which gave them control over the integration APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition → on both sides of the connection. If you are running a third-party MES alongside SAP S/4HANA, you are negotiating with two vendor roadmaps simultaneously, and that is a governance problem before it is a technology problem.

Closing on the practical point: the Amberg architecture works because someone made a decision about who owns which data object and enforced it. No amount of middleware sophistication substitutes for that decision being made and written down. CDOs who inherit a plant with two systems and no data ownership mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → should start there, before touching the integration layer.

The full course on this sector:Data in Manufacturing.