Can you rely on the Treasury Department’s Sanctions List Search?November 11, 2014
When the United States wants to prohibit its citizens and corporations from doing business with a foreign national, that individual is added to the Specially Designated Nationals list maintained by the Office of Foreign Assets Control of the US Department of the Treasury. One person on that list is Chabaane Ben Mohamed al-Trabelsi*, a Tunisian associated with Al-Qaeda who has lived in Italy.
So if you try to wire money to someone named “Chabaane Ben Mohamed al-Trabelsi” the transaction will likely be flagged for further review. In that further review additional information (e.g., date of birth) would be used to determine if he is the terror funder. But what if you use a variant of that name like “Shab’anah Bin-Muhammad Tarabulsi”? You can see this name differs from the name above in spelling, punctuation and the use of the Arabic definite article “al”. How would the bank make sure this variant name was flagged for further review?
That’s where name matching software comes in to make sure any similar name, not just an exact match, is flagged for review. Our Rosette software is used by a number of institutions for exactly this purpose.
Recently, the US Department of the Treasury’s Office of Foreign Assets Control developed Sanctions List Search (SLS). SLS is “designed to facilitate the use of the Specially Designated Nationals and Blocked Persons list (‘SDN List’) and all other non-SDN lists.” According to the SLS web site it:
“uses approximate string matching to identify possible matches between word or character strings as entered into Sanctions List Search, and any name or name component as it appears on the SDN List and/or the various non-SDN lists …. Sanctions List Search will detect certain misspellings or other incorrectly entered text, and will return near, or proximate, matches”
We saw SLS as a perfect opportunity to compare the accuracy of and flexibility of Rosette with the analogous tool that the US Government is providing as a resource. The results were a bit disconcerting – but before we get to them, let’s talk about how we got them.
From AKAs to Accuracy Tests
A person’s entry on the SDN list can also include ‘also known as’ names or AKAs. These AKAs are often just known variants of the official name that the keepers of the list add to try to make sure they’re found.
Realizing this, we wondered what would happen if the AKAs weren’t on the list. What if the keepers of the SDN list hadn’t added a specific AKA, which was used in issuing a wire transfer, would it still be able to be found? Would different matchers still be able to find the person being mentioned if none of the AKAs were present? These aren’t just hypothetical questions because the list of AKAs can grow over time.
With this testing approach defined, we returned to the individual on the list mentioned above, Chabaane Ben Mohamed al-Trabelsi. His entry includes a number of AKAs including the variant mentioned above, “Shab’anah Bin-Muhammad Tarabulsi”. We wanted to know if SLS and Rosette find his official name when queried with this variant:
|Query: “Shab’anah Bin-Muhammad Tarabulsi”|
|Result #||Sanctions List Search||Rosette|
|1||Asil Sami Mohamad Madhi Tabrah||Chabaane Ben Mohamed Trabelsi|
|2||Ali Sa’d Muhammad Mustafa Bakri||[no other matches above threshold]|
|3||Al-Azhar Ben Ammar Ben Abadallah Tlili|
|4||Husam Muhammad Amin Al-Yassin|
|5||Chabaane Ben Mohamed Trabelsi|
SLS incorrectly identifies the names of 4 other people that it feels are a better match than the name that we know is correct. That’s a significant amount of false positives, leading to a precision of just 20%. With Rosette, only the correct name is returned. Rosette finds the other incorrect names that SLS does but with a score well below the threshold for returning a match.
This is just one of the names we ran through our comparative analysis. We discovered numerous cases where the SLS produces multiple false positives as compared to Rosette, as well as false negatives, where the correct name falls below the match threshold on the SLS system, but is identified as a match when using Rosette.
SLS vs. Rosette
So with the tool provided by the Department of Treasury, either you need a rich inventory of AKAs to help find matches or plenty of time to wade through the false positives. We would be concerned if SLS were used by regulators or other officials as a benchmark for high-accuracy fuzzy name matching. In contrast, we believe Rosette represents the state-of-the-art in high-accuracy fuzzy name matching, whether for automatic or interactive applications.
*This individual has been de-listed by the United Nations Security Council Al-Qaeda Sanctions Committee and was acquitted of all terrorism-related charges by the Milan courts. An application to remove him from the SDN list is currently pending.