ETL, ELT and workflow automation: what is the difference?

ETL: extract, transform, load

ETL reads data from a source, reshapes it, and writes the finished result to a destination. The transformation happens before loading, so the destination only ever receives clean data.

It suits situations where the destination should not hold raw data — a reporting database, a file sent to a partner, a system of record with strict rules about what it accepts.

ELT: extract, load, transform

ELT loads raw data first and transforms it inside the destination, usually a data warehouse with plenty of compute. The raw data stays available, which helps when requirements change.

It suits analytical work on large volumes, where keeping the original is more valuable than keeping the destination tidy.

Workflow automation

Workflow automation is broader. It is about the sequence of steps in a process — which may include moving data, but also producing a document, calling an API, waiting for an approval or notifying someone.

The distinction matters less than it used to. Most real processes are a mixture: read some data, clean it, decide something, produce an output and tell someone.

What to look for in a tool

Whatever the label, a few properties decide whether a pipeline survives contact with real work.

• You can see the data at every step before anything is written.
• The preview and the real run use the same engine, so they cannot disagree.
• A failed run says what failed and stops before writing half a result.
• Someone other than the author can read the pipeline and understand it.
• It handles the awkward inputs — PDFs, Word documents, spreadsheets with merged headers — not only tidy tables.