ELT stands for extract, load, transform. You pull data out of your sources, load it into a central store in its raw form, and only then shape it into the clean tables your reports use. The transform step happens inside the warehouse, not on a separate server along the way.
Picture moving house. The old way (ETL) is unpacking and sorting every box in the van before you carry anything inside. ELT is the opposite: you carry every box in first, then sort through them in the new house where you have space and tools. Because cloud data warehouses became fast and cheap, doing the sorting inside the warehouse often beats doing it on the road. The raw boxes stay around too, so you can re-sort later if you change your mind.
That is the real advantage over ETL: the raw data is preserved, so when a new question comes up you rebuild from what is already loaded instead of going back to the original system. The trade-off is that you need a warehouse strong enough to handle the transform work itself.
In practice the transform step is usually written in SQL and managed by a tool like dbt, which lets analysts shape the data without a separate engineering team. A common stack pairs Fivetran for the load with BigQuery for the store. The marketing team gets a fresh dashboard, and nobody touched a line of pipeline code to add a new field.
At TopDevs we lean on ELT when a client runs a modern cloud warehouse, so their raw data stays intact and new reports can be built without re-importing anything.
