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

Extraction pulls raw records from wherever they live — databases, APIs, files, event streams — often on a schedule. Transformation is where most of the engineering effort goes: cleaning malformed values, deduplicating records, converting types and units, joining data from multiple sources, and applying business rules so the output is trustworthy and queryable. Loading writes the transformed result into its destination, typically in batches, and the whole pipeline is usually run on a schedule or triggered by upstream events. ETL's counterpart, ELT, reverses the order — load raw data first, transform inside the powerful destination warehouse — which has become more common as cloud warehouses got cheaper to compute in. Either way, the core value is the same: turning fragmented, inconsistent source data into something a downstream system, report, or AI agent can rely on without needing to know the quirks of every origin system. Pipeline reliability (retries, schema-change handling, data quality checks) matters as much as the transformation logic itself.