Extracting, loading, and transforming (ELT) data are central to the two most common data engineering processes. First, data engineering services must always extract data in some way from a single source, but what happens next is not so straightforward.

ELTs are common in Data lake stores or architectures that require raw extracted data from multiple sources. It enables different processes and systems to process data from a single extraction. When combining data from various sources, it is advantageous to co-locate and store the data in a single location before converting it.

Meanwhile, the ETL (extract, transform, load) process heavily computes the conversion before loading the results into a file system, database, or data warehouse. Compared to the above process, it is not as efficient because each batch or stream frequently requests data from dependent systems. It means you have to re-query data from them, add load to those systems, and wait for data to become available on each execution.

However, in cases where simple transformations are applicable for a single data source, ETL may work better because it reduces system complexity at the expense of data exchangeability.