How to Design a Medallion Architecture for Manufacturing Data: A Practical Guide

Manufacturers generate enormous amounts of data every second. ERP systems, MES platforms, SCADA systems, PLCs, IoT sensors, warehouse systems, quality applications, maintenance software, CRM platforms and finance systems all produce valuable information.

The trouble is that most of that data sits in silos. Production teams work in MES. Finance depends on the ERP. Operations run on spreadsheets. Maintenance uses a CMMS. Executives see reports that are already a day late. The result is inconsistent KPIs, poor visibility and slow decisions.

This is where Medallion Architecture has become the preferred data-engineering approach for modern manufacturing. Whether you are implementing Microsoft Fabric, Azure Databricks, Snowflake or another lakehouse platform, it provides a scalable framework for turning raw manufacturing data into trusted business intelligence.

Medallion Architecture is a layered approach to data engineering, where data improves in quality as it moves through three stages:

• Bronze layer — raw data
• Silver layer — validated and clean data
• Gold layer — business-ready data

Instead of transforming data the moment it lands, the pattern preserves the original and enriches it gradually. That gives you better governance, easier troubleshooting and analytics that scale.

A typical manufacturer manages data from dozens of operational systems — SAP S/4HANA, Microsoft Dynamics 365, Oracle ERP, Siemens MES, Rockwell FactoryTalk, Ignition SCADA, PLC controllers, warehouse management systems, Salesforce, quality management systems, maintenance systems, IoT devices and plenty of Excel files.

Each system stores data its own way: different naming conventions, different timestamps, duplicate records, missing values and master data that keeps changing. Without a proper architecture, reporting becomes unreliable.

A Medallion Architecture creates a single trusted foundation for analytics on top of all of it.

The Bronze Layer stores data exactly as it arrives. Nothing is modified, cleaned or deleted. It is your permanent landing zone.

This gives you:

• A complete audit trail
• Historical backup
• Data lineage
• Replay capability
• Compliance

Typical manufacturing sources:

• ERP transactions and production orders
• Machine telemetry and sensor data
• Maintenance logs
• Purchase orders and supplier deliveries
• Inventory movements and warehouse transactions
• Financial journals

Data formats often include CSV, JSON, XML, SQL tables, REST APIs, and MQTT or OPC-UA streams.

The Silver Layer turns raw data into trusted datasets. This is where the engineering effort goes.

Data cleansing — remove duplicate records, invalid values, corrupted rows and missing timestamps.

Standardisation — reconcile the same thing recorded differently. Plant A logs WH01, Plant B logs Warehouse-01, Plant C logs Main Warehouse; all become Warehouse 01.

Master data integration — join Product, Customer, Supplier, Plant, Equipment and Employee masters so every record carries consistent context.

Business rules — apply the logic once, here, not in every report. For example: machine runtime = stop time − start time; actual production = good units + scrap; inventory value = quantity × standard cost; plus OEE availability, shift and production-day calculations.

Data quality checks — flag missing product IDs, invalid machine IDs, negative production, future timestamps and duplicate orders. Only trusted data progresses to Gold.

The Gold Layer holds business models optimised for reporting and AI — the datasets your dashboards and models actually consume.

Executive KPIs — revenue, gross margin, EBITDA, inventory value, working capital.

Manufacturing KPIs — OEE, availability, performance, quality, throughput, yield, scrap rate, cycle time, downtime.

Supply chain KPIs — OTIF, DIFOT, supplier performance, lead time, inventory turns, stock ageing.

Quality KPIs — first pass yield, defect rate, NCR trends, CAPA performance.

Maintenance KPIs — MTBF, MTTR, equipment availability, planned-maintenance compliance.

These datasets become the source for Power BI, Tableau and Excel, and for AI, machine learning and Copilot on executive dashboards.

The data moves in one direction, improving at each step:

• Sources — ERP, MES, SCADA, PLC, IoT sensors, CRM, WMS, CMMS, finance
• Bronze layer — raw data
• Silver layer — validated data
• Gold layer — business models
• Consumption — Power BI, AI, machine learning, executive dashboards, predictive analytics

Organise data by business domain, not by source system — domain-based design scales far better than an application-by-application structure.

1. Never modify Bronze data. Raw data should stay immutable. Always.

2. Keep business logic out of reports. Avoid embedding calculations in Power BI. Business rules belong in Silver or Gold, so every report uses the same logic.

3. Use incremental data loads. Loading 500 million records a day is wasteful. Load only new, updated and — where relevant — deleted records; it cuts processing time sharply.

4. Monitor data quality continuously. Track missing records, duplicates, failed pipelines, invalid master data and late-arriving data — not just at go-live.

5. Use Delta tables or similar ACID formats. Transactional storage gives faster queries, time travel, version history, schema evolution and reliable updates.

6. Build reusable data models. Instead of separate logic per dashboard, build reusable semantic models that serve multiple departments.

Most manufacturers hit the same pitfalls:

Reporting straight from the ERP. ERP systems are built for transactions, not enterprise analytics.

Mixing raw and clean data. Never let reporting tools consume Bronze data.

Hardcoding business rules. Logic changes; centralise it in Silver or Gold so maintenance stays simple.

Ignoring historical data. Trend analysis, forecasting and AI need years of production history.

Department-specific data silos. Every team should read from the same trusted foundation — one source of truth keeps KPIs consistent.

Several platforms support the pattern well. The architecture stays the same regardless of the technology underneath.

Manufacturers that implement Medallion Architecture well typically see:

• Faster reporting cycles and decisions
• Improved data quality and consistent enterprise KPIs
• Better governance and compliance
• Less reporting maintenance and lower engineering cost over time
• Scalable AI and machine-learning initiatives
• Simpler integration of new data sources
• Greater confidence in the numbers

Manufacturing leaders are under pressure to decide faster. But dashboards and AI are only as good as the data beneath them.

A well-designed Medallion Architecture turns fragmented operational data into a trusted, governed analytics platform built to scale. Structuring data into Bronze (raw), Silver (validated) and Gold (business-ready) layers supports everything from operational reporting and executive dashboards to predictive maintenance, demand forecasting and AI-driven optimisation.

Whether you build on Microsoft Fabric, Azure Databricks or another lakehouse platform, investing in the pattern now creates the data foundation your smart-manufacturing plans will depend on.

At MyData Insights, we design and implement data platforms that unify ERP, MES, IoT, SCADA, warehouse, finance and quality systems into a modern lakehouse — Microsoft Fabric and OneLake, Azure Databricks and Delta Lake, end-to-end Medallion Architecture, manufacturing KPI and semantic-model development, Power BI executive dashboards, AI and predictive analytics, and data governance.