Name Matching (Fuzzy Matching)

Names are poor identifiers. The same person can appear as Mohammed, Muhammad or Mohamad, with or without a middle name, written in Arabic, Cyrillic or Latin letters. An exact comparison would miss most of these. Systems first apply name normalisation: they strip punctuation, titles and company suffixes, ignore capital letters, and sort name parts into a standard order. Transliteration tables then map characters from one script into another so that a name written in a foreign alphabet can be compared with its Latin form. The fuzzy step measures similarity with methods such as edit distance, sound-alike codes and comparison of name parts, and returns a score. A threshold decides which scores become alerts. Set it low and analysts drown in false alerts; set it high and genuine matches slip through. Tuning that trade-off, and proving it works, is the usual regulatory question. Weak matching undermines every other screening control, however good the lists are.

An insurer’s onboarding system screens a new policyholder named Alexandr Petrov against a list holding Aleksandr Petrova. Exact matching finds nothing. After normalisation and transliteration, the fuzzy score comes to 92 out of 100, above the insurer’s alert threshold of 85. An analyst checks the date of birth and country, sees that they differ, and clears the alert with a recorded reason.