- Extract: Data is extracted from various source systems, such as databases, APIs, or files. The data can come from relational databases, flat files, cloud applications, or other structured/unstructured sources.
- Load: Instead of transforming the data before loading, the raw data is first loaded into the target system, such as a cloud data warehouse, data lake, or distributed database (e.g., Azure Synapse, BigQuery, Snowflake).
- Transform: The transformation step is done after loading, within the target system itself. Since modern cloud platforms and distributed databases have strong computing power, they handle large volumes of data and complex transformations more efficiently.
ELT is preferred for cloud-based architectures and big data platforms where transformation can be done “on-demand” and the target system can store large amounts of raw data. ELT takes advantage of the scalability and processing power of modern cloud data warehouses.
See also: ETL vs ETL