Medallion architecture: best practices for managing Bronze, Silver and Gold

Piethein Strengholt
10 min readJan 13, 2023

Many of my clients employ a Medallion structure to logically arrange data in a Lakehouse. They process incoming data through various stages or layers. The most recognized layout, illustrated below, incorporates Bronze, Silver, and Gold layers, thus the term “Medallion architecture” is used.

Although the 3-layered design is common and well-known, I have witnessed many discussions on the scope, purpose, and best practices on each of these layers. I also observe that there’s a huge difference between theory and practice. So, let me share my personal reflection on how the layering of your data architecture should be implemented.

Data platform strategy

The first and most important consideration for layering your architecture is determining how your data platform is used. A centralized and shared data platform is expected to have quite a different structure than a federated multi-platform structure that is used by many domains. The layering also varies based on whether you align platform(s) with the source-system side or consuming side of your architecture. A source-system aligned platform is usually easier to standardize in terms of layering and structure than a consumer-aligned platform given the more diverse data usage characteristics on the consumption side.

With these considerations in mind, let’s explore each layer after each layer. For each layer, I first provide some abstract and high-level objectives. After that, I’ll make the layering more specific with observations from the field.

Landing area

A landing area, or landing zone, is an optional level frequently implemented by organizations setting up a data platform. It serves as a temporary storage location for data gathered from various sources before it is transferred into the Bronze layer. This layer becomes particularly necessary when extracting data from the target source system proves challenging, such as when dealing with external clients or SaaS vendors. In these instances, there may be a dependency, or data might be received in an unsuitable file format or structure.



Piethein Strengholt

Hands-on data management professional. Working @Microsoft.