ETL vs ELT: what's the difference?

ETL and ELT both move data from a source to a destination, but they order the transform step differently. ETL transforms data before loading it, shaping it to fit the destination first. ELT loads data raw, then transforms it in place inside the destination. Cloud warehouses made ELT the common pattern once storage and compute got cheap.

Same steps, different order

Both patterns extract data from a source and land it in a destination. The letters describe the order transform happens in relative to load: ETL does extract, transform, load, so the destination never sees unshaped data. ELT does extract, load, transform, so a full raw copy exists before any cleaning runs.

Why ELT won out in the cloud era

ETL made sense when warehouse storage was expensive and compute for transforming data lived on a separate server. Once cloud object storage got cheap and query engines got fast enough to transform data where it sits, keeping a raw copy stopped costing much, and having it around for reprocessing or auditing turned out to matter.

Which one OptimaFlo uses

OptimaFlo runs ELT. Raw data lands in the Raw layer exactly as the source sent it, zero transforms, full history, under Iceberg's ACID guarantees. The Clean layer then applies LLM-generated SQL against that already-landed data, and Ready builds business metrics on top of Clean. Nothing gets reshaped before it lands.

Related terms

• ETL pipeline: The transform-before-load pattern this page compares ELT against.
• Apache Iceberg: The table format that makes OptimaFlo's raw Raw-layer landing safe to keep and query.