Skip to main content

Fuzzy Matching Algorithms

Last Updated: 2025-12-28 Status: Complete

This page covers the algorithms and techniques used for name matching in sanctions screening.

Why Exact Matching Isn't Enough​

Relying on exact name matches would allow sanctions evasion through:

  • Typos and misspellings: "Mohammad" vs. "Muhammad"
  • Transliteration variations: Arabic/Cyrillic names to Latin alphabet
  • Name order differences: "John Smith" vs. "Smith, John"
  • Nicknames and aliases: "Bob" vs. "Robert"
  • Missing middle names or initials
  • Corporate name variations: "ABC Corp" vs. "ABC Corporation" vs. "ABC Inc."

Fuzzy matching algorithms calculate similarity scores between names to identify potential matches even when spelling differs.

Algorithm Comparison​

AlgorithmMethodBest ForSpeedUsed By
Levenshtein DistanceCounts minimum edit operations (insert/delete/substitute) to transform one string to anotherComplex names, business namesSlowerMost commercial AML solutions
Jaro-WinklerCharacter similarity with prefix weighting (rewards matching first characters)Individual names, typosFasterOFAC Sanctions List Search tool
SoundexPhonetic encoding (encodes names by how they sound)Similar-sounding namesVery fastOFAC Sanctions List Search tool
Double MetaphoneAdvanced phonetic encoding with multiple encodings per nameComplex pronunciationsFastAdvanced screening systems

Levenshtein Distance​

Calculates the minimum number of single-character edits (insertions, deletions, substitutions) required to change one string into another.

Example:

  • "Mohammed" → "Muhammad" = 3 edits (o→u, e→a, d→d is same)
  • Lower distance = higher similarity

Strengths:

  • Handles typos well
  • Works for any character set
  • Intuitive interpretation

Weaknesses:

  • Slower for long strings
  • Doesn't account for phonetic similarity

Jaro-Winkler​

Measures character-level similarity with a boost for matching prefixes (first 4 characters).

Example:

  • "SMITH" vs "SMYTH" = 0.96 similarity (high match)
  • Names with same first letters score higher

Strengths:

  • Fast calculation
  • Good for names where first letters are reliable
  • Standard in OFAC tools

Weaknesses:

  • Prefix bias may miss some variations
  • Less effective for names with different start

Soundex​

Converts names to phonetic codes based on sound, then compares codes.

Example:

  • "Smith" → S530
  • "Smyth" → S530
  • Same code = potential match

Strengths:

  • Very fast
  • Handles phonetic variations
  • Good for English names

Weaknesses:

  • Limited for non-English names
  • Coarse matching (high false positives)

Double Metaphone​

Advanced phonetic encoding that generates multiple codes per name to handle pronunciation variations.

Example:

  • "Schmidt" → ["XMT", "SMT"]
  • Matches both German and Anglicized pronunciations

Strengths:

  • Handles international names better
  • Multiple encoding options
  • More accurate than Soundex

Weaknesses:

  • More complex implementation
  • Still primarily English-focused

OFAC's Official Approach​

The OFAC Sanctions List Search tool uses fuzzy matching algorithms (commonly reported to include Jaro-Winkler and Soundex-based approaches) to identify potential matches even when spelling differs.

Key Points:

  • OFAC's tool is for reference, not compliance
  • Commercial systems typically use multiple algorithms
  • PayFacs should not rely solely on OFAC's search tool

Setting Match Thresholds​

Common Threshold Ranges​

  • 95-100%: Very likely match - investigate immediately
  • 85-94%: Possible match - manual review required
  • 70-84%: Lower probability - may review based on risk appetite
  • Below 70%: Typically ignored (but configurable)

Risk-Based Threshold Setting​

  • High-risk industries: Lower thresholds (more sensitive)
  • High-risk geographies: Lower thresholds
  • Common names: May need higher thresholds (e.g., "Mohammed Ahmed")
  • Business names: Often need lower thresholds (more variation)

Threshold Tuning Process​

  1. Start conservative: Lower thresholds catch more potential matches
  2. Analyze results: Track false positive rates by name type
  3. Adjust incrementally: Raise thresholds for high false-positive categories
  4. Document changes: Maintain audit trail of threshold decisions
  5. Review quarterly: Re-evaluate based on new data
Threshold Tuning is Critical

Initial threshold settings will generate thousands of false positives. Continuous tuning based on review outcomes is essential to balance compliance with operational efficiency.

Multi-Algorithm Approach​

Best practice is to combine multiple algorithms:

Score = (0.4 × Levenshtein) + (0.35 × Jaro-Winkler) + (0.25 × Soundex)

Benefits:

  • Catches different types of variations
  • Reduces false negatives
  • More robust matching

Implementation:

  • Most commercial vendors use proprietary combinations
  • Tune weights based on your merchant population
  • Test with known sanctions list entries
Share: