Which approach is most efficient when performing a fuzzy match against a master list of 10 million records to improve performance?