Thomson Reuters Uses AI to Turn World-Check Tool into Financial Robocop

Thomson Reuters has rolled out an upgraded media screening component within its World-Check One financial crime and risk monitoring platform, using artificial intelligence to filter potentially relevant negative mentions in news and text data relating to companies or individuals, allowing analysts to more quickly perform risk assessments and decide whether a counterparty could represent undue business risk.

The new tool, Media Check, aggregates 11,600 print and online media sources, and allows users to search for individuals, companies or vessels to help identify potential instances or links to money laundering, theft, fraud, cybercrime, and political exposed persons. Media Check applies Thomson Reuters Intelligent Tagging—proprietary machine learning-based algorithms that process and tag unstructured content with a “unique” financial crime-based taxonomy.

“AI through machine learning handles large volumes of unstructured data, identifies entities, content themes, and forms relationships between a theme and an entity. Therefore, it allows greater precision when searching for content related to an entity,” says Mike Moreno, senior product manager for Risk at Thomson Reuters. “We had a media screening solution in the first version of World-Check One, but it was quite basic, and clients still got too many false positives and duplicates, so there was a need to upgrade it.”

Therefore, the vendor began working with a small number of clients who had reported issues to collect new requirements for an improved version.

As the vendor analyzed these issues, Moreno’s team realized that “A lot of the problems being raised in Risk had already been raised elsewhere within Thomson Reuters—for example, in Westlaw [Thomson Reuters’ legal database],” he says. “So we went shopping.”

At the end of its spree, the team managed to pull in “de-duping” technology—which eliminates duplicate occurrences of the same item from different sources—from its Eikon financial workstation, and expanded on the existing capabilities to include a “clustering” feature that groups relevant information together, such as creating connections and relationships between names and articles. Media Check clients will also have access to WestLaw’s NewsRoom feature, which aggregates both licensed and web-crawled news from different publishers beyond the vendor’s own Reuters News operations.

“De-duping will reduce complexity around the information, while the Intelligent Tagging will reduce false positives,” Moreno says.

AI performs three key tasks within the new Media Check: first, de-duping; second, enabling clustering; and third, establishing “entity risk relationships”—i.e., intelligently connecting words in a sentence, rather than just basing a judgement on how closely they are grouped. For example, in the past, attorney names would have inevitably had negative associations because they would be closely linked to court cases or arrests, even if they weren’t the defendant’s attorney, and weren’t the individual being arrested. So in the past, these would have always given false positive results, Likewise, a search on a police officer’s name might reveal negative connotations because they could be linked to multiple arrests, even though they were the arresting officer, rather than the individual being arrested for a crime. By using AI and machine learning, Media Check will be able to better understand the nature of those relationships and deliver more accurate results.

Thomson Reuters began an early adopter phase with a “minimum viable product” version of Media Check in mid-April to elicit feedback from potential users. Moreno says the vendor will decide the end date for the early adopter testing when it feels comfortable with the product.