What is an ELT (Extract, Load, Transform)?

This differs from the older ETL pattern (extract, transform, load) mainly in where and when transformation happens. ETL transforms data in a separate processing layer before it ever reaches its destination, which historically kept destination systems cheap but made every schema change a slower, riskier pipeline edit. ELT became dominant once cloud data warehouses got cheap and powerful enough to run heavy transformations themselves — loading raw data fast and iterating on transformation logic (often as version-controlled SQL) inside the warehouse. The practical benefits are faster iteration (change a transformation without re-running the whole extraction), a full raw-data audit trail (you still have the untransformed source data if a transformation bug is discovered later), and simpler pipelines since the load step doesn't need to know the final schema in advance. ELT is now the default pattern in most modern data stacks, typically implemented with a connector tool for extraction/loading and a SQL-based transformation layer on top.