It's used whenever you need to quickly get a representative sample of data from a large data set for testing purposes.

It's especially useful when you're developing data warehouses as you can develop and test quickly without having to process a very large database every time you make an amendment to a report or cube.