ELT
ELT (Extract, Load, Transform) is a data integration pattern that loads raw data into a destination system first and performs transformation afterward, using the destination's own compute resources.
Definition
ELT (Extract, Load, Transform) is a data integration pattern that loads raw data into a destination system first and performs transformation afterward, using the destination's own compute resources.
Overview
ELT reorders the classic ETL pipeline: Extract still pulls data from source systems, but Load happens immediately after, depositing raw or lightly processed data directly into the destination — typically a cloud data warehouse or data lakehouse — before any significant transformation occurs. Transform then runs inside that destination system, using its own compute power to clean, join, and reshape the data via SQL. This pattern became practical largely because of the shift to cloud data warehouses like Snowflake, BigQuery, and Amazon Redshift, which decouple storage from compute and can scale transformation workloads elastically, something on-premises warehouses historically couldn't do cost-effectively. Because raw data lands in the warehouse untransformed, teams retain the ability to reprocess or re-derive different transformed views later without re-extracting from the original source system — a significant advantage when transformation logic or business requirements change. dbt (data build tool) is the tool most closely associated with the modern ELT pattern: it doesn't extract or load data itself, but manages the Transform step entirely as version-controlled SQL that runs against already-loaded raw data, with dependency management, testing, and documentation built in. Extraction and loading are typically handled by separate tools — Fivetran, Airbyte, or custom scripts orchestrated by Apache Airflow — that feed raw data into the warehouse for dbt to transform. ELT has become the dominant pattern for modern analytics engineering teams, though it assumes the destination system can handle both the extra storage of raw data and the additional transformation compute load, a trade-off worth understanding alongside traditional ETL when designing a data platform, as explored in dbt & Analytics Engineering.
Key Concepts
- Loads raw data into the destination before any significant transformation
- Transformation runs inside the destination using its own compute
- Enabled by decoupled storage/compute architecture of modern cloud warehouses
- Preserves the ability to reprocess raw data with new transformation logic
- dbt is the leading tool for managing the Transform step in ELT pipelines
- Extraction/loading typically handled by separate tools (Fivetran, Airbyte, Airflow)
- Now the dominant pattern for modern analytics engineering teams