ETL vs ELT
Extract-Transform-Load and Extract-Load-Transform differ in whether data is reshaped before or after it lands in the destination store.
Two orderings of the same verbs
Both patterns extract data from a source, transform it into a usable shape, and load it into a destination. The difference is order. In ETL, transformation happens in a dedicated processing stage before the data reaches its home. In ELT, raw data is loaded first and transformed inside the destination system, often a warehouse with strong compute.
ETL
ETL suits cases where the destination is expensive to compute in, where raw data must be cleaned before it is allowed to land, or where privacy rules require stripping fields early. The transform step runs in a separate engine, so the warehouse only ever sees the finished product.
- Pro: only clean, conformed data lands; smaller storage footprint.
- Pro: sensitive fields can be dropped before persistence.
- Con: reprocessing needs re-extraction; the raw form may be lost.
- Con: transformation logic is separated from the query engine.
ELT
ELT loads raw data first, then transforms it with the destination's own SQL or compute. It keeps the original bytes, so new transformations can be derived later without touching the source. Modern columnar warehouses and lakehouses made ELT practical by making in-place transformation fast.
- Pro: raw data is preserved for replay and new derivations.
- Pro: transformation logic lives beside the data as versioned queries.
- Con: raw storage grows; governance must cover unfiltered data.
- Con: sensitive fields exist in the store until scrubbed.
Choosing
Scientific programs often prefer ELT for the raw archive: the original measurement is the ground truth and must never be discarded, and derived quantities are recomputed as calibration improves. A separate ETL path may still feed a curated, privacy-safe subset to public consumers. See data lakes and reproducible datasets.