How Fuzzy Matching Improves OFAC Screening Accuracy
The simplest way to screen a name against the SDN list is an exact match: does this string appear on the list, yes or no? It's fast and easy to understand. It also misses a lot.
Why Exact Matching Falls Short
Sanctioned individuals and entities often appear under multiple name variations:
- Transliteration differences. Arabic, Cyrillic, and Chinese names can be transliterated to English in multiple valid ways. "Muammar Gaddafi" has at least 30 documented English spellings.
- Name order. Some cultures put the family name first. "Kim Jong Un" might appear as "Jong Un Kim" in a Western system.
- Abbreviations and initials. "Mohammed" becomes "M." or "Mohd" or "Moh'd."
- Typos in your own data. A misspelled name in your customer database won't match the correctly spelled SDN entry.
An exact-match search would miss all of these. Fuzzy matching is designed to catch them.
How Fuzzy Matching Works
Edit Distance (Levenshtein)
This measures how many single-character changes (insertions, deletions, substitutions) are needed to turn one string into another. "Smith" to "Smyth" is one substitution, so the edit distance is 1. Low edit distance means the names are similar.
Phonetic Algorithms
Algorithms like Soundex and Metaphone convert names into phonetic codes based on how they sound in English. "Smith" and "Smyth" produce the same code. This catches names that sound alike but are spelled differently.
Token-Based Comparison
Names are split into tokens (individual words) and compared regardless of order. This means "Al-Rashid, Mohammed" matches "Mohammed Al Rashid" even though the order and punctuation differ.
Combined Scoring
In practice, screening tools combine these techniques and produce a confidence score, typically 0 to 100. A score of 100 means an exact match. A score of 85 might mean the names are very similar with minor spelling differences. A score of 60 might be a partial match worth reviewing.
Setting Your Threshold
The threshold is the minimum score at which a match gets flagged for review. Lower thresholds catch more potential matches but generate more false positives. Higher thresholds reduce noise but risk missing real matches.
There's no single right answer. Most compliance teams settle in the 75-85 range after some tuning. Start lower, review the results for a few weeks, and adjust based on the volume and quality of matches you're seeing.