Why exact matching fails in sanctions screening

Name matching is the core technical challenge in sanctions screening. Unlike exact database lookups, screening must account for the myriad ways a single name can be recorded, transliterated, misspelled, or abbreviated. The US Treasury explicitly acknowledges that sanctions compliance requires more than exact matching - a recognition that the gap between how names appear in business records and how they appear on sanctions lists is significant and systematic.

The scale of name variation

One person can appear under many spellings. Three sources of variation cause most missed matches.

Transliteration challenges

Arabic, Chinese, Russian, and other non-Latin script names can have dozens of valid romanised forms. Research published in Language Resources and Evaluation (Springer) demonstrates that a single Arabic name can produce over 30 different transliteration variants. The name "محمد" alone can appear as Muhammad, Mohammed, Mohamed, Mohamad, Muhammed, and many other forms - all linguistically valid.

Cultural naming conventions

Different cultures follow different naming conventions that confound simple matching approaches. East Asian names may reverse family and given name order; Hispanic names use dual surnames; some cultures use patronymics rather than fixed family names. Without understanding these patterns, screening systems generate excessive false positives or, worse, miss genuine matches.

Aliases, titles, and abbreviations

Sanctioned individuals frequently operate under aliases, nicknames, or abbreviated forms of their legal names. OFAC's SDN List includes alternative names ("also known as" entries), but business records may capture yet another variant not listed as an official alias.

Algorithmic approaches to fuzzy matching

Fuzzy matching techniques fall into a few families, and each catches a different kind of variation.

String similarity algorithms

Levenshtein distance measures the minimum number of single-character edits (insertions, deletions, substitutions) needed to transform one string into another. It captures typographical errors and minor spelling variations effectively.

Jaro-Winkler similarity gives additional weight to prefix matches, reflecting the empirical observation that the beginnings of names are less likely to vary than endings. This makes it particularly effective for Western naming conventions where given names tend to remain stable.

Phonetic algorithms

Soundex encodes names by their phonetic sound in English, grouping together names that sound alike. Double Metaphone extends this to handle a wider range of European and non-European phonetic patterns. These algorithms catch names that sound similar but are spelled very differently - for example, "Schmidt" and "Smith" share phonetic roots.

Hybrid and machine learning approaches

A comprehensive review in Expert Systems with Applications found that hybrid approaches combining string similarity, phonetic matching, and machine learning significantly outperform any single technique. Modern engines weight multiple signals - name similarity, date of birth, nationality, country of residence - to produce a composite confidence score.

Balancing sensitivity and false positives

Every matching system trades missed matches against false alarms. Getting the balance wrong is costly in either direction.

The cost of false positives

Every false positive alert requires manual review by a compliance officer. For law firms and accounting practices with limited compliance resources, excessive false positives create operational bottlenecks and alert fatigue - a condition where genuine matches risk being overlooked amid a flood of false alerts.

Configurable thresholds

FirmScreen's configurable matching engine allows compliance teams to adjust sensitivity weights for name similarity, date of birth, country, and nationality. Higher confidence scores indicate stronger matches, allowing teams to prioritise investigation effort effectively. A firm screening low-risk domestic clients may set a higher threshold than one handling cross-border transactions.

Threshold configuration by risk profile

  • High sensitivity (70%+): Recommended for crypto and VASP compliance where regulatory tolerance for misses is low
  • Standard sensitivity (80%+): Appropriate for most DNFBP use cases including real estate and legal
  • Precision mode (90%+): Suitable for high-volume batch rescreening where minimising review workload is priority

Practical implications for compliance teams

Understanding how fuzzy matching works enables compliance officers to make better-informed decisions about match results. When reviewing an alert, knowing whether the match was triggered by string similarity, phonetic matching, or date-of-birth correlation helps assess its credibility. FirmScreen provides transparent match scoring that breaks down the contributing factors for each alert.

For batch screening workflows, see our guide on batch vs real-time screening approaches.