ELT stands for extract, load, transform. Data is extracted from sources and loaded raw into the data platform first, then transformed there with the platform's own compute, usually in SQL. It replaced the older ETL order in most cloud platforms because storage is cheap and the platform scales the transformation work.
For example, an ELT pipeline for a web shop might copy the raw orders, customers and products tables from the shop's database into bronze tables every hour, unchanged. SQL running on the platform then cleans them into silver tables, removing duplicates and fixing types, and joins them into a gold table of daily revenue by product category. If the revenue definition changes next month, the team rewrites the SQL and rebuilds gold from the raw data already loaded, without extracting anything again.
In the older ETL order, data was transformed in a separate tool before it was loaded, and only the transformed result reached the warehouse. That made sense when warehouse storage and compute were expensive. It also meant raw data was often not kept, so a change in business rules required going back to the source, which might no longer have the history.
ELT moves the risk rather than removing it. Raw data in the platform can include personal or sensitive values, so access to bronze needs care, and transformation code becomes the place where business logic lives, so it needs tests, review and orchestration like any other code. See also medallion architecture, SQL model and lakehouse.
In practice, the extract and load steps are often handled by a connector or ingestion tool, while the transform step lives in SQL files in a repository, built in order by the platform's scheduler. Keeping those two halves separate means a change to a business rule never needs a change to the ingestion code, and the other way around. See also change data capture.
Emolysis follows the ELT order: Ingest (batch) loads raw data to bronze, and later shapes transform it on your platform. Pushing results back out is the separate Reverse ETL shape.