ETL vs ELT
Same three letters, different order, and it changes your whole data platform.
The idea
ETL transforms data before loading it: clean it in Python or a tool, then load only the result. ELT loads the raw data first and transforms it inside the warehouse with SQL.
ELT took over as storage got cheap and warehouses got powerful: keeping the raw data means you can rerun new logic without extracting again. ETL still wins when sensitive data must be cleaned or masked before it lands anywhere.
What the lesson will build
- The same messy orders data run both ways
- ETL with pandas, ELT with SQL inside DuckDB
- A side-by-side of what each one keeps
Key ideas
- Extract, transform, load
- Raw versus modeled data
- Masking PII before it lands
- Where dbt fits
The video
- Long form: Two pipelines stacked, with the T block sliding to a new position and a PII column masked in one lane but not the other.
- Short: Transform first, or load first. Here’s when each wins.
The lesson is ready: code and a line-by-line walkthrough in data-engineering/12-etl-vs-elt/.