Many organisations have invested in a modern data platform, yet still struggle to deliver consistent, reliable insights. Data arrives from multiple systems, moves through different pipelines, and often lands in formats that are difficult to trust or use.
Medallion architecture offers a structured way to manage this. It applies a layered approach to data lakehouses, where data is progressively refined as it moves through clearly defined stages. This makes it easier to organise data pipelines, improve data quality, and support both technical teams and the business user.
The Bronze, Silver, and Gold layers support a more dependable approach to data engineering and data transformation.
Start from the top: What is Lakehouse Architecture?
What is Medallion Architecture?
Medallion architecture is a data design pattern used to organise data within a data lakehouse medallion architecture. It structures data into layers that reflect its level of refinement, from raw ingestion through to business-ready outputs.
The pattern introduces a layered data model:
- Bronze for raw data
- Silver for cleaned and standardised data
- Gold for curated, consumption-ready data
This layered approach creates a clear pathway for transforming data. Each stage has a defined purpose, which helps teams manage moving data between systems while maintaining consistency.
Within a modern data platform, this structure supports:
- Stronger data management practices
- More predictable data pipelines
- Better alignment between data engineering and business intelligence teams
Make sure you understand the difference: Data Lakes vs. Data Warehouses: Which is Best for Your Business?
The Bronze Layer: Raw, Immutable Data
The Bronze layer is the entry point of the data lake medallion architecture. It captures data in its original form, directly from source systems.
This layer focuses on ingestion rather than transformation. Data is typically:
- Loaded in bulk or streamed in near real time
- Stored with minimal changes
- Appended rather than updated
The goal is to preserve a complete and accurate record of source data. This supports traceability and allows teams to reprocess data if needed.
In practical terms, the Bronze layer plays a key role in moving data from operational systems into the data lakehouse. It often includes data from:
- ERP and CRM systems
- APIs and third-party platforms
- IoT devices and event streams
Schema handling is usually flexible at this stage. Some teams apply light structure on ingestion, while others rely on schema-on-read. The priority is to land data quickly and reliably.
For example, a retail organisation may ingest point-of-sale and online transaction data into the Bronze layer. This ensures a complete record of sales activity, which can be revisited for auditing or reprocessing as business rules change.
Working with multiple data sources? Unify them into a single system: Data Lakehouse Architecture: The Best of Lakes and Warehouses
The Silver Layer: Cleaned and Conformed Data
The Silver layer is where data transformation becomes more deliberate. Data is cleaned, validated, and aligned across sources to improve usability.
Typical processing in this layer includes:
- Removing duplicates
- Handling missing or incorrect values
- Standardising formats and naming conventions
This is also where schema enforcement is applied. Unlike the Bronze layer, the Silver layer introduces structure and consistency, which helps improve data quality across the platform.
A key responsibility of the Silver layer is to bring related datasets together. Customer data from multiple systems can be matched and standardised into a consistent format, creating a reliable foundation for downstream use.
For example, a financial services provider may use the Silver layer to reconcile customer records from legacy systems. By applying consistent rules and schema enforcement, they can produce a unified dataset that supports compliance reporting and analytics.
The Gold Layer: Business-Ready Data Products
The Gold layer is where data is shaped for consumption. It represents the final stage of the data medallion architecture, where datasets are refined into forms that support reporting, analytics, and decision-making.
At this level, data is:
- Aggregated and modelled around business concepts
- Structured for business intelligence tools
- Designed with the business user in mind
Rather than focusing on technical structure, the Gold layer reflects how the organisation operates. Data is often organised into domain-specific data products, such as sales performance, customer insights, or operational metrics.
This layer plays a key role in establishing a single source of truth. By standardising definitions and calculations, teams can rely on consistent metrics across reports and dashboards.
For example, an energy provider may build Gold datasets that track electricity demand by region and time. These datasets support forecasting models and executive reporting, helping teams respond to changes in usage patterns.
Are you already using AI and data services? Make sure you’re getting the max ROI: The Business Value of AI and Data Services in Modern Enterprises
How Delta Lake Enables Medallion Architecture
Delta Lake provides the foundation for implementing medallion architecture within a modern data platform. It adds reliability and structure to data stored in a data lake, making it suitable for production-grade data pipelines.
Key capabilities include:
- ACID transactions, which ensure consistent reads and writes
- Versioning, allowing teams to track and roll back changes
- Time travel, enabling access to historical versions of data
These features support each stage of the data lakehouse medallion architecture. For example:
- Bronze data can be ingested continuously without risking corruption
- Silver transformations can run reliably with consistent inputs
- Gold datasets can be updated with confidence and traceability
Delta Lake also simplifies pipeline design. Instead of managing complex workarounds, teams can rely on built-in mechanisms for handling updates, merges, and streaming data.
Schema Evolution and Enforcement in Practice
As data moves through a layered data model, structure becomes more important. Two concepts are central here: schema enforcement and schema evolution.
Schema enforcement ensures that data meets defined rules. It prevents unexpected changes from breaking downstream processes, which is particularly important in the Silver and Gold layers.
Schema evolution allows those structures to change in a controlled way. This is necessary because source systems rarely stay the same. New fields are added, formats shift, and requirements change over time.
In practice, this balance supports reliable data engineering:
- New columns can be introduced without disrupting existing models
- Data types can be updated with proper validation
- Pipelines can adapt without needing full redesign
Within tools like Delta Lake, schema evolution is handled alongside versioning. This makes it easier to manage change while maintaining consistency across the data lake medallion architecture.
Incremental Processing: Scaling Efficient Data Pipelines
Processing all data from scratch is rarely practical at scale. Incremental processing addresses this by only handling new or changed data as it moves through the pipeline.
This approach is a key part of efficient data transformation. It reduces processing time, lowers costs, and supports more responsive analytics.
In a medallion architecture, incremental processing fits naturally into each layer:
- Bronze: Data is appended as it arrives from source systems
- Silver: Transformations process only new or updated records
- Gold: Aggregations are refreshed based on changes rather than full rebuilds
Data build tools (DBT) are commonly used to manage this within a medallion architecture DBT setup. They allow teams to define incremental models that update datasets in a controlled and repeatable way.
This method keeps data pipelines efficient while supporting near real-time insights. It also helps maintain a scalable data platform, where growth in data volume does not lead to a proportional increase in processing overhead.
Learn the basics of Databricks before getting into data layers: What is Databricks? The Data Intelligence Platform Explained
Medallion Architecture in Modern Platforms
The strength of medallion architecture is that it is not tied to a single toolset. It is a data design pattern that can be applied across different technologies, depending on how a data platform is structured.
In a Databricks environment, the pattern is closely aligned with Delta Lake. Tables are organised into Bronze, Silver, and Gold layers, with built-in support for versioning and incremental updates.
In Microsoft ecosystems, the MS Fabric medallion architecture follows a similar structure. Data is ingested into OneLake and refined through staged transformations, supporting both engineering workflows and business intelligence use cases.
For teams using DBT, the pattern is expressed through models and transformations. A medallion architecture DBT approach typically maps staging models to Bronze, intermediate models to Silver, and marts to Gold. This keeps data transformation logic modular and easier to maintain.
Despite differences in tooling, the intent remains consistent:
- Separate raw and curated data
- Apply structure progressively
- Deliver trusted datasets for consumption
Gain the Value of a Layered Approach to Data
Medallion architecture introduces discipline into how data is handled over time. Instead of treating ingestion, transformation, and reporting as separate concerns, it connects them through a consistent structure.
That structure creates space for better decisions. Teams spend less time reconciling differences and more time applying insight where it counts. For organisations investing in data lakehouses, this approach provides a steady foundation.
Idea 11 works alongside organisations to design and implement data lakehouse medallion architecture that fits their environment and priorities. From ingestion through to curated data products, our approach is practical, adaptable, and grounded in real delivery experience.
Discover how we can support your goals with a structured, scalable approach to data.
