Layered Architecture for Analytics Engineers
Layered architecture organises transformations into stages: staging models clean raw sources, intermediate models combine logic, and marts expose business-ready tables. Patterns such as medallion layers, one big table and data vault offer other trade-offs between flexibility, speed and ease of use.
Stage: Data Modelling · Part of the Analytics Engineer roadmap · Last reviewed 19 November 2026
What to learn in Layered Architecture
- Create staging models that rename and cast raw columns only
- Place joins and business logic in intermediate models
- Build a mart designed for a specific business team
- Compare medallion layering with staging, intermediate and marts
- Decide when a one big table is acceptable
Layered Architecture interview questions
What belongs in a staging model?
A staging model sits one-to-one with a source table and does light cleaning only: renaming columns, casting types, standardising values and basic deduplication. Joins and business rules are kept for later layers so staging stays simple and reusable.
What are the trade-offs of a one big table?
A wide denormalised table is fast and easy for analysts to query, but it duplicates data, can be costly to rebuild and hides grain. It suits specific dashboards, while shared models usually stay normalised.
Where Layered Architecture fits in the Analytics Engineer roadmap
Layered Architecture belongs to the Data Modelling stage (intermediate level), which takes about 5 hours. Structure data for people and machines.
Other units in this stage
Back to the full Analytics Engineer roadmap