ELT (extract, load, transform)

ELT is a data integration pattern in which raw data is extracted from source systems and loaded directly into a destination, usually a cloud data warehouse or data lakehouse, with transformation happening afterward inside that destination using its own compute engine.

The order of operations is what separates it from ETL, where transformation happens before loading on a separate processing tier. ELT became practical once cloud warehouses made storage cheap and compute elastic, so it is now efficient to load everything first and reshape it with SQL. Tools like dbt popularized this approach by letting analytics engineers write transformations as version-controlled SQL models that run against already-loaded raw data, making the modeling layer transparent and testable rather than hidden inside an integration tool.

ELT matters because it decouples ingestion from transformation, letting raw data land quickly and giving analysts room to iterate on models without re-running extraction. It also keeps an untransformed copy of source data available for reprocessing. The main pitfall is treating "load everything, model later" as a substitute for governance: without clear ownership of the transformation layer, raw and modeled data blur together and downstream data quality suffers.