Put simply, the point of data sampling ine-Discovery is to let you learn something about a large body of ESI without spending a lot of money. It is a great tool to answer questions like:
1) How good a job did my review team do to identify privileged documents? (QC check)
2) How many responsive documents are likely to be on this backup tape?
3) Of the documents I received ine-Discovery, what percent are REALLY responsive?
4) How often does a specific search term produce a useful document?
And lots of similar questions that might impact what you do next, and how costly a case might end up being.
How does Sampling work, and how much money can I save using it?
First, you have to decide how precise you want your conclusions to be. Obviously if you want to be 100% confident and have a zero margin of error, you need to check every single document, and sampling is of no use for you.