What Is ELT? Extract, Load, Transform Explained
ELT stands for Extract, Load, Transform: load raw data into the warehouse first, then transform it in place using the warehouse engine.
ELT reorders the classic ETL steps to fit modern cloud warehouses. Instead of transforming data before loading, ELT loads raw data straight into a powerful warehouse and runs transformations there, using SQL and tools like dbt.
What ELT is
ELT extracts data from sources, loads it raw into the warehouse, then transforms it inside the warehouse. The transform step runs as SQL against scalable cloud compute, so the warehouse does the heavy lifting rather than a separate processing tier.
Why the order changed
Cloud warehouses made compute cheap and elastic, so it became practical to load everything raw and transform on demand. This keeps the raw data available for re-processing if requirements change, and lets analysts build transformations in SQL without a separate engine.
The role of dbt
ELT popularized tools like dbt, which manage the transform step as version-controlled, tested SQL models built in dependency order inside the warehouse. The extract-and-load step is often handled by a connector tool like Fivetran or Airbyte.
ELT in CI
Because transformations are SQL models in version control, CI can build and test them against a temporary schema before they merge.
steps:
- run: dbt build --select state:modified+ --target ci
- run: dbt test --target ciLatchkey note
ELT CI jobs are SQL-heavy and network-bound rather than compute-bound. On Latchkey, auto-retry helps a long dbt build survive a warehouse-connection blip, and caching the toolchain keeps these frequent pull-request jobs starting quickly.
Key takeaways
- ELT loads raw data into the warehouse first, then transforms it in place with the warehouse engine.
- Cheap, elastic cloud compute made loading raw and transforming on demand practical.
- Transformations live as version-controlled SQL models and are built and tested in CI before merging.