Extract, Transform, Load

ETL, short for extract, transform, load, is a data pipeline pattern that pulls data out of source systems, reshapes it into a consistent form, and writes it into a destination such as a data warehouse. The extract step reads from databases, APIs, log files, or SaaS applications. The transform step cleans values, reconciles differing schemas, removes duplicates, and applies business rules. The load step writes the result where analysts and downstream applications can query it. Running this on a schedule gives reporting a stable, unified view instead of a scatter of raw systems. Airflow and Dagster orchestrate the steps, Fivetran and Airbyte handle extraction from common sources, and Snowflake or BigQuery often serve as the destination. Many teams now invert the middle two steps into ELT: load raw data first, then transform it inside the warehouse with SQL, an approach dbt popularized. Cheap warehouse storage and elastic compute made that ordering practical. The main pitfall is silent breakage. When a source renames a field or changes a type, a pipeline can keep running while producing wrong numbers, so pipelines need schema checks, data tests, and monitoring rather than only success and failure alerts.