Transaction Screening Applies to Both Structured and Unstructured Data
In today’s highly regulated financial environment, financial institutions are required to screen transactions against various watchlists for risk management, fraud detection, and compliance with Anti-Money Laundering (AML) and Counter-Terrorism Financing (CTF) regulations.
These watchlists are issued by government agencies, such as the Office of Foreign Assets Control (OFAC) within the U.S. Department of the Treasury and by law enforcement, and include specific individuals and entities that have been involved in money laundering, terrorism financing, criminal activities, etc. Failure to comply can result in severe financial penalties and reputational damage for financial institutions.
While the main transaction details come as structured data (e.g., sender, beneficiary), regulatory requirements also apply to the unstructured or free-text fields in transaction messages (e.g., SWIFT, credit card transactions, bank transactions). The challenge is that these free text fields are short and telegraphic in nature, with frequent use of abbreviations, acronyms, and codes, and may be hard to interpret for the uninitiated.
Entity Extraction Can Help
Free-text transaction fields may contain names of individuals and organizations that need to be matched against the watch lists. These names may occur mixed with other things that are not part of the names. A necessary step is to apply a technology, Entity Extraction, that can identify the names prior to matching them against the watch lists.
Every day, financial institutions process millions of short text strings in free-text transaction fields such as “POS 1234 WALGREENS SEATTLE WA” or “AMZN Mktp US*AB12C3.” Entity Extraction identifies the component pieces of such strings, enabling high-volume processing that would be much, much faster than a human team.
Entity Extraction identifies items such as:
- Person names
- Organization names
- Location names, particularly cities and countries
- Vessels
- Crypto wallet addresses
- Etc.
Without effective and real-time Entity Extraction, much of this information found in free-text transaction fields would remain inaccessible without enormous human effort and increased compliance risk for financial institutions.
How Entity Extraction Works
Entity Extraction, an AI technology, automatically identifies key concepts such as names in unstructured text and classifies them into predefined categories, thereby creating structured data.
Entity Extraction doesn’t just identify concepts. It knows what kind of thing the concept is:
- John Robinson – person name
- XYZ Corporation – company
- Russia – country.
It does this not by having a long list of names. Instead, it uses AI technology that recognizes the characteristics of a name and so can make the determination that a name is present.
Entity Extraction recognizes names dynamically – its most significant contribution is that it identifies names that have not been seen before. This dynamic recognition allows it to recognize names well beyond the capabilities of a necessarily limited look-up list.
Advanced Entity Extraction is robust with respect to grammatical irregularities that are omnipresent in free-text transaction fields. Some entity extraction software still relies on things like capital letters on initial letters in a name to recognize the presence of a name. Advanced Entity Extraction can recognize names even when their initial letters are not capitalized or are incomplete or misspelled.
Here are some examples of what Entity Extraction extracts when applied to free-text transaction fields:
ATM Deposit-CASH App*Bill Roscoe
ENTITY_TYPE: PERSON
NAME: Bill Roscoe
XYZ CORP PAY DAVID REMINGTON
ENTITY_TYPE: COMPANY
NAME: XYZ CORP
ENTITY_TYPE: PERSON
NAME: DAVID REMINGTON
ACH – Majestic Motors SETTLEMENT Horizon
ENTITY_TYPE: COMPANY
NAME: Majestic Motors
ENTITY_TYPE: COMPANY
NAME: Horizon
From unstructured text, Entity Extraction produces structured data of entity names and entity types. Once they are extracted, they can then be matched against whatever watch lists the financial institution is using for sanctions screening.
Summary
Entity Extraction transforms raw free-text transaction fields into structured data. By identifying entity names and types in the text, financial institutions can exploit this additional data to match against sanctions lists and achieve much more complete sanctions screening.
Entity Extraction supports real-time transaction monitoring despite ever-growing transaction volumes and continually updated sanctions lists. As a result, robust Entity Extraction is a critical capability for accurate transaction screening.
