You can use partition exchange to swap a staging table with a partition in a fact table. Loading directly into the fact table can be slow performance because:

• There can be a large number of indexes on the fact table (slowing inserts)
• If these are bitmap indexes, this limits your ability to load in parallel (bitmap indexes can block concurrent DML)

You can avoid some of these issues by loading the data into a staging table. If you have a fact table partitioned by (transaction) day, then you can:

• Drop indexes on the staging table
• Load the data in
• Create indexes and gather stats (on the staging table)
• Exchange the staging table with the target partition in the main fact table.

The exchange partition is near-instant and will load the data complete with indexes. When working with large data sets, this process can make the ETL significantly faster.

If you'd like more details, have a look at Tim Gorman's "Scaling to Infinity" article.