Dealing with fuzzy matching false positives in sanctions screening

par :name Aisha Rahman · Conformité aux sanctions · Mar 20, 2026 · 3 réponses Répondu

Aucune garantie sur le contenu du forum. Les informations, opinions et discussions partagées sur ce forum sont fournies par les membres de la communauté et l'équipe LexFlag et ne constituent pas des conseils professionnels. LexFlag n'approuve, ne vérifie ni ne garantit l'exactitude, l'exhaustivité ou la fiabilité du contenu publié.

Identité des utilisateurs et contenu généré par l'IA. Rien ne garantit que les utilisateurs utilisent leur vrai nom, représentent une organisation ou expriment leurs propres opinions. Les réponses et contributions peuvent être partiellement ou entièrement générées par l'intelligence artificielle.

Vérification indépendante requise. Vous devez vérifier de manière indépendante toute information obtenue sur ce forum avant de prendre toute décision. LexFlag, ses affiliés et les contributeurs déclinent toute responsabilité pour toute perte ou tout dommage résultant de la confiance accordée au contenu du forum.

Our sanctions screening system uses fuzzy matching with a threshold of 85%. This catches legitimate matches but also generates thousands of false positives per month, especially for common names.

Examples of problematic patterns:

"Mohammed Ali" matching against dozens of sanctioned individuals
Common Chinese names generating excessive hits
Partial name matches inflating results

How do you balance catch rate (not missing true matches) with operational efficiency? What matching algorithms and thresholds have you found to be optimal?

Aisha Rahman

Chief Compliance Officer · NeoBank Digital

Membre depuis Apr 2026

Réponse acceptée

This is the eternal sanctions screening dilemma. A few approaches that helped us:

Secondary filters — After the initial fuzzy match, apply secondary criteria: date of birth, nationality, gender. This alone eliminated ~40% of our false positives.
Name normalization — Implement transliteration rules for Arabic, Chinese, and Cyrillic names. Many false positives come from inconsistent romanization.
Whitelisting — For repeatedly cleared entities, maintain a whitelist with periodic revalidation. Our regulator was fine with this as long as we documented the initial clearance rationale.
Multiple algorithms — We use a combination of Jaro-Winkler, phonetic matching (Soundex), and exact match. Different algorithms catch different types of matches.

We settled on an 80% threshold with secondary filters, which reduced false positives by about 55% without increasing false negatives.

John Doe

Compliance Officer · Lexonica Inc.

Membre depuis Mar 2026

3 réponses

False positives in sanctions screening are the #1 operational pain point we hear about. A few strategies beyond threshold tuning:

Whitelisting with guardrails — Build a vetted whitelist of known-good entities that repeatedly trigger false matches. But put an expiration date on each entry (6-12 months) and require revalidation. Whitelists that never get reviewed become compliance gaps.

Name normalization before matching — Standardize diacritics, transliterations, common abbreviations (Ltd/Limited, Corp/Corporation), and honorifics before the matching algorithm runs. A significant percentage of false positives come from format differences rather than actual name similarity.

Secondary data filtering — If your initial screen returns a potential match, automatically compare dates of birth, nationality, and address data before surfacing it to an analyst. This can auto-clear 30-40% of false positives without human intervention.

Feedback loops — Track which types of alerts are consistently resolved as false positives and use that data to adjust your matching configuration. If "Mohammed" + "Ali" generates 200 alerts per month that are all cleared, your system needs a contextual rule for that name combination.

The industry average false positive rate for sanctions screening is 95-98%. Getting that below 90% with good tuning is achievable without increasing your miss rate.

LexFlag Team

Mar 21, 2026 at 11:33 AM

One underrated technique: context-aware matching. Instead of just matching the name, also look at related data points. If "Mohammed Ali" shows up as a hit, but the customer is a 25-year-old from Germany and the sanctioned individual is a 60-year-old from Syria, you can quickly disposition that.

Some modern screening platforms automate this contextual comparison, which significantly reduces the manual review burden.

John Matcher

Mar 23, 2026 at 9:33 AM