How Do Banks Use AI to Screen Customers Against Sanctions Lists?

Why Sanctions Screening Is Legally Required

Banks are legally required to screen customers and transactions against government-maintained sanctions lists, which identify individuals, entities, and in some cases entire countries that are prohibited or restricted from certain financial dealings. In the U.S., the Treasury Department’s Office of Foreign Assets Control (OFAC) maintains the primary sanctions lists that financial institutions must check against. Failing to catch a genuine sanctions match can expose a bank to significant legal and regulatory consequences, which is why sanctions screening is treated as a core, non-negotiable part of compliance rather than an optional risk management practice.

Why Simple Name Matching Doesn’t Work Well Enough

A naive approach to sanctions screening might involve simply checking whether a customer’s exact name appears on a sanctions list. In practice, this approach misses far too much to be reliable. Names get transliterated differently from non-Latin alphabets (a single name written in Arabic or Cyrillic script, for instance, can have multiple valid English spellings), people use nicknames or aliases, and typos or minor formatting differences between how a name appears in a bank’s system versus on a sanctions list are common. A bad actor attempting to evade detection might also deliberately use a slightly altered version of their name.

AI-based approaches address this using fuzzy matching techniques, which identify likely matches based on similarity rather than requiring an exact character-for-character match, and natural language processing methods that can better handle name variations, transliteration patterns, and contextual information like date of birth or nationality that helps distinguish a genuine match from an unrelated person who happens to share a similar name.

Managing the Trade-Off Between Missed Matches and False Positives

Sanctions screening involves a similar fundamental trade-off to other AML systems: a screening approach that’s too loose risks missing genuine matches, which carries serious legal and safety consequences, while an approach that’s too broad generates excessive false positives, flagging large numbers of unrelated people who simply share a common name with someone on a sanctions list. AI-based systems aim to strike a better balance than simpler matching approaches by weighing additional contextual information, like date of birth, nationality, or address, alongside name similarity, to more accurately assess whether a potential match is likely genuine.

Regardless of how sophisticated the automated matching is, potential matches are generally routed to human compliance analysts for investigation and confirmation before any account action is taken, and sanctions lists themselves are updated regularly, requiring banks to continuously rescreen their existing customer base, not just check new customers at account opening.

Bottom Line

Banks use AI-based fuzzy matching and natural language processing to screen customers against sanctions lists like OFAC’s, catching name variations and transliterations that simple exact-match searches would miss, while routing potential matches to human compliance analysts for final investigation and confirmation.

Go deeper

• What Was the Mata v. Avianca AI Sanctions Case About?