What is medallion data architecture?
Medallion architecture is a data platform pattern that progressively refines data as it moves through the platform.
Bronze stores data as it arrives from source systems.
Silver cleans, validates and combines data into consistent records.
Gold prepares data for reporting, analytics, machine learning and operational use.
Databricks describes the pattern as a way to improve data structure and quality step by step across a lakehouse. Google Cloud uses the same three-layer logic to explain how raw data becomes curated information for specific business needs.
The layers are logical boundaries. They don't always need to be separate physical copies of every dataset. A small reference table might use a view, while a large analytical dataset may require materialized tables at each stage.
Medallion architecture provides a technical structure. The operating model still needs to define who owns each dataset, who approves changes and how quality is measured.
The three layers of medallion architecture
Bronze layer: raw source data
The Bronze layer is the landing area for data from operational databases, SaaS applications, APIs, files, event streams and other source systems.
The goal is to preserve the source record with as little change as possible. Teams usually add technical metadata such as:
Ingestion timestamp
Source system
File name or event identifier
Batch or process ID
Load status
Schema version
Bronze data gives engineering teams a reliable starting point for reprocessing. If a transformation rule changes, they can rebuild Silver data from the original records instead of requesting the data again from the source system. The layer also supports audit trails and lineage because it preserves what entered the platform and when.
Bronze still needs controls. Basic checks should confirm that files arrived, records are within an expected range and schema changes have been detected. Invalid records can be sent to a quarantine area rather than silently discarded.
Access should be restricted because Bronze may contain raw personal, financial or operational data. Platform engineers and ingestion services usually need access. Most business users should work with Silver or Gold datasets instead.
Silver layer: validated and conformed data
The Silver layer turns raw records into data that teams can use across several analytical workloads.
Typical transformations include:
Data type conversion
Deduplication
Null handling
Standardized dates, currencies and codes
Schema validation
Reference-data joins
Identity resolution
Change data capture processing
Masking or removal of sensitive fields
Silver data should retain meaningful detail. It is usually more general than a dashboard dataset, which makes it useful for analysts, data scientists and downstream data products.
This layer is also where teams create consistent views of business entities such as customers, products, employees, suppliers or transactions. Data from several source systems can be joined and conformed, giving the organization a shared structure for further analysis.
Silver needs the strongest data quality checks in the pipeline. A customer dataset might require unique customer IDs, valid country codes and consistent contact fields. A financial dataset might need reconciled totals, valid accounting periods and complete transaction references.
Records that fail validation should be rejected, quarantined or marked with a clear status. The pipeline should record the reason, the affected dataset and the team responsible for resolution.
Gold layer: business-ready data
The Gold layer contains datasets designed for a specific business purpose.
Examples include:
Monthly revenue by region
Customer lifetime value
Inventory availability
Sales pipeline performance
Workforce reporting
Churn model features
Executive dashboards
Gold datasets usually apply business rules, aggregations and final calculations. They may use star schemas, denormalized tables, data marts or feature tables. The design depends on how users will query the data and which decisions the dataset needs to support.
A Gold dataset should have a clear description, named owner, refresh frequency, quality status and approved use cases. Users need to know which metric definitions apply and when the data was last updated.
Gold is where trust becomes visible to the business. A dashboard that shows revenue, margin or customer activity needs documented calculations and a clear route back to its source data.
Bronze, Silver and Gold compared
Layer | Main purpose | Typical users | Main quality gate | Ownership focus |
|---|---|---|---|---|
Bronze | Preserve source data and technical history | Data engineers, platform teams, audit teams | Arrival, schema and ingestion checks | Source and pipeline ownership |
Silver | Clean, validate and standardize detailed data | Engineers, analysts, data scientists | Completeness, accuracy, consistency and referential checks | Domain data ownership |
Gold | Deliver business-ready datasets and metrics | Business users, BI teams, executives, ML teams | Business rules, reconciliation and freshness | Product or use-case ownership |
The same dataset can support several Gold outputs. For example, a Silver customer dataset might feed a marketing dashboard, a finance report and a machine learning feature table. Each Gold output can have a different owner and quality target.
Quality gates between the layers
The boundaries between layers should work as quality gates.
The Bronze to Silver gate checks whether raw records can be trusted for wider use. It can test schema, data types, required fields, duplicate keys, reference values and record counts.
The Silver to Gold gate checks whether the data supports a specific business purpose. It can test metric calculations, reconciliation with source totals, freshness targets, aggregation logic and policy requirements.
Each gate should define:
The conditions data must meet.
The team that owns the check.
The action taken when the check fails.
The information recorded in the catalog or pipeline logs.
The process for correcting and replaying failed data.
A failed check should stop or flag the affected output. Publishing a dashboard with incomplete sales data creates a business problem that is harder to trace later.
Governance responsibilities across the layers
Medallion architecture supports governance when responsibilities are attached to each layer.
Ownership
Every important dataset needs an accountable owner. Ownership can sit with a central data team, a business domain or a product team, depending on the target architecture.
The owner approves definitions, sets quality expectations and decides how changes are managed. Technical teams can operate pipelines, but they shouldn't have to decide what a business metric means.
Metadata and lineage
Each dataset should include metadata such as:
Business definition
Source system
Data owner
Update frequency
Sensitivity classification
Retention rule
Quality status
Downstream dependencies
Lineage should show how data moves from Bronze to Silver to Gold. This helps teams investigate quality incidents, assess the impact of source changes and explain how a reported metric was calculated. ER/Studio also identifies lineage and validation as important governance concerns within a medallion design.
Access and protection
Access should reflect the sensitivity and purpose of each layer.
Bronze may require strict access because it contains raw records. Silver can provide broader access after validation and masking. Gold can support wider business use when sensitive details have been removed, aggregated or protected.
Access rules should be applied through roles, policies and regular reviews. The platform should record who accessed sensitive datasets and which transformations were applied.
Cost and retention
Keeping every dataset forever at every layer increases storage and processing costs. Retention policies should reflect the purpose of each layer.
Bronze may need longer retention for replay, audit or regulatory reasons. Silver may retain detailed history for operational and analytical use. Gold can retain only the period required for reporting, forecasting or business analysis.
Benefits and limits of the pattern
A well-designed medallion architecture can give teams:
A repeatable path from raw data to business use
Better traceability when data quality problems occur
Reusable Silver datasets for several analytical products
Clearer access rules by data maturity and sensitivity
Faster query performance through purpose-built Gold datasets
A shared vocabulary for platform and business teams
The pattern also creates practical risks.
Gold sprawl happens when every dashboard creates its own version of a metric. The result is conflicting revenue, customer or performance figures. Gold datasets should be linked to approved definitions and business owners.
Skipped Silver layers create repeated cleaning logic. If each Gold table reads directly from raw data, teams may implement deduplication and business rules differently across reports.
Unclear ownership turns the layers into technical folders. A platform team may maintain the pipelines while nobody owns the meaning, quality or business use of the data.
How to implement medallion architecture
A practical implementation can follow 6 steps.
Start with a priority data domain. Choose a business area such as customers, finance or supply chain. Define the decisions and use cases the platform must support.
Map the data flow. Identify source systems, Bronze objects, Silver transformations, Gold outputs, consumers and dependencies.
Write layer contracts. Define what each layer contains, which transformations are allowed and what must be true before data crosses into the next layer.
Assign ownership. Name the source owner, dataset owner, pipeline owner and Gold product owner. Record decision rights for definitions and changes.
Automate quality checks. Add validation to ingestion and transformation pipelines. Store quality results with the dataset metadata.
Monitor use and failure. Track freshness, failed checks, unresolved issues, access requests, processing costs and the number of certified Gold datasets.
From layers to a governed platform
Medallion data architecture gives a lakehouse a clear path from raw records to business-ready information. Bronze preserves the source, Silver creates consistent data and Gold supports specific decisions.
The pattern creates value when every layer has a purpose, an owner and a measurable quality standard. It should fit the organization’s target architecture and operating model rather than dictate them.
Organizations designing their data platform, governance model or ownership structure can explore data strategy consulting.
