If you were given a dataset with millions of records, how would you approach analyzing it?