ETL stands for extract, transform, load: the process of pulling data out of source systems, reshaping it, and writing it into a destination such as a data warehouse. Extraction reads from databases, files, or APIs. Transformation cleans, joins, and restructures the data. Loading writes the result where analysts and applications can use it.
ETL contrasts with ELT, which loads raw data first and transforms it inside the destination using its compute. Modern cloud warehouses made ELT popular, but ETL remains standard where data must be validated or reshaped before it lands.
For AI teams, ETL determines what the model actually sees. The same source data can produce different training sets depending on transformation logic and timing, so recording which pipeline version and which data state produced each output keeps results reproducible.
Related terms: ELT · Data Pipeline · Data Warehouse · Data Ingestion