Compliance

Sanctions Name Screening Matcher

Compare a customer list against a watchlist you supply using Jaro-Winkler, token and phonetic matching, with a tunable threshold and a report of which technique produced each hit.

Last reviewed by the Radiatus Cloud team

Results appear here.

Going for ISO 27001, SOC 2, HIPAA or GDPR?

Radiatus runs end-to-end compliance & GRC programs.

Get a free readiness review

Exact matching fails on the names it matters most for

Sanctions and watchlist screening breaks down on transliteration long before it breaks down on anything else. A name written in Arabic, Cyrillic or Chinese script has no single correct romanisation, so the same person legitimately appears as Mohammed, Muhammad, Mohamad and Mohamed. Exact matching misses all but one of those, which is why every real screening system is fuzzy and why the interesting question is not whether to use fuzzy matching but where to set the threshold.

The threshold is a business decision disguised as a parameter

Lowering it catches more true matches and produces more false positives, and the ratio is not linear: below about 0.85 on most similarity measures the false positive count rises far faster than the true positive count. A screening team that cannot clear its alert queue will start clearing alerts carelessly, which is worse than a slightly higher threshold honestly set. The right threshold is the one that produces a queue the team can actually work through.

Different techniques catch different failures

Edit-distance measures catch typographical errors and small transliteration differences. Token matching catches reordering and dropped middle names, which defeat edit distance entirely because Smith John and John Smith are far apart character by character. Phonetic matching catches names that sound alike but are spelled differently. Running all three and reporting which one fired tells the reviewer what kind of similarity they are looking at, which is most of what they need to clear or escalate the alert.

Related tools

Frequently Asked Questions

Does this tool include a sanctions list?

No, and that matters. Watchlists change constantly and an embedded copy would be out of date and unsafe to rely on. You supply the list, which means you control its currency and provenance.

What threshold should I use?

Most programmes sit between 0.85 and 0.92 on a Jaro-Winkler scale. Below about 0.85 the false positive count rises far faster than the true positive count, and a queue the team cannot clear gets cleared carelessly.

Why run three matching techniques?

Because they catch different failures. Edit distance catches typos, token matching catches reordering and dropped middle names, and phonetic matching catches names that sound alike but are spelled differently. Reporting which one fired tells the reviewer what they are looking at.

Is a high score a confirmed match?

No. A score is a similarity measurement, not an identification. Confirming a match requires secondary identifiers such as date of birth, nationality or an identification number, none of which name similarity can supply.

Is anything sent to a server?

No. All comparison runs in your browser, and neither list is transmitted. That is deliberate, since both lists are usually sensitive.

Privacy & Security

Everything runs in your browser; nothing is uploaded.

Data: None
Client-side-Side
Active
v1.0

How to Use

Paste your names and the watchlist to compare against.

Disclaimer: This tool is provided "as is" without warranty of any kind. Results are for educational and utility purposes.