Cleaning first and only reaching for edit distance for what is left is a sensible order, and the warning that first initial plus last word gives false positives is honest. One thing about the 0.85 cut-off: NAMESCORE divides by the longer length, so short names cross it easily. 'Ann Lee' against 'Ann Lea' is one edit over 7 characters, about 0.86, and 'Jon Smith' against 'Jan Smith' is about 0.89. Both pairs are different people. For names I would make the cut-off depend on length, or send anything short that scores below 1 to a 'check by eye' column instead of accepting it. It also helps to feed the sorted key from Step 1 into NAMESCORE rather than the raw cell. Otherwise 'Smith, John' against 'John Smith' scores very low, even though the cleaning step already knows they are the same name.