Can a computer learn to speak trader?

Illustration: Thomas R. Lechleiter/WSJ, iStock
Illustration: Thomas R. Lechleiter/WSJ, iStock

Summary

Compliance software firms are pushing artificial intelligence to decode Wall Street’s near-impenetrable jargon.

Microsoft and Google can translate, say, Spanish to English, but who can translate finance speak?

Up-and-coming compliance software firms are leaning into artificial intelligence to help crack esoteric trader argot, as Wall Street and London watchdogs increasingly dive into traders’ jargon-loaded and even coded messages to help fight financial crime.

Financial conspiracies are often built on the clandestine communications between financial professionals in which they agree to, for example, rig an interest-rate benchmark or increase the price at which a currency is trading.

Even an educated lay person would need an insider to explain the lingo used between traders. But compliance software companies are rushing to provide firms with AI tools that enable their systems to outsmart crooked traders.

Compliance staff can use a tool from Behavox to input a chat between two traders and receive an explanation of what their slang means and whether they are potentially trying to do something illicit.

Behavox, a London-based compliance software company that sells to large banks, hedge funds and other firms, offers tools that can ingest the huge amount of messages that financial firm employees generate and look for the financial crime needle-in-the-haystack of jargon-laden, everyday work.

Compliance staff can use the Behavox tool to input a chat between two traders and receive an explanation of what their slang means and whether they are potentially trying to do something illicit. The language models the tool uses were trained almost the way a student would be trained, by exposing them to regulatory websites, securities filings and other sources to take them beyond the subject-matter depth of a generalist AI tool, said Chris Bauer, the head of technical services at Behavox.

Behavox can also translate from other languages so that an English-speaking compliance staffer in New York, for instance, could monitor a Japanese speaker in Tokyo.

“We’ve tested this on hundreds of different languages. A co-worker of mine swears he’s got it working on Klingon. I’m not 100% sure," said Bauer, nodding to the alien language from the “Star Trek" movie and television series. “The bottom line is that we have the ability to take in different not just literal translations, but also jargon as well."

Minding the naughty list

The particulars of financial industry record-keeping requirements have evolved as the ability to capture and store certain communications has improved with technological innovation. Wall Street regulators first required recording of at least some telephone calls in 1998 under what is known as the “Taping Rule."

U.S. regulators now treat essentially any written communication regarding business as a record. They have levied billions in fines against firms for failing to rein in traders who talked shop on messaging services like WhatsApp.

Until recently, a common way for companies to handle sifting through the enormous quantity of emails that a Wall Street firm produces even in a day was to apply what is known as a lexicon, an ever-growing list of words and phrases that trigger human review.

One problem: Any phrase can be innocuous or illicit, depending largely on context, so the lexicon approach can result in a large number of false positives.

Lexicons can also produce false negatives, as traders adapt their communications to try to avoid trigger words. A trader might write “Ca11 m3 n0w," or turn to emojis to try to duck detection.

Global Relay, another compliance firm that serves banks and other financial industry clients, is testing a new AI system with customers that can quickly adapt to approaches that would defeat a traditional tool, said Donald McElligott, the vice president of compliance supervision at the company. The company says its pilot system has been able to decipher phrases coded with emojis, and even pig Latin.

Traders believed that “if somebody wanted to say something sketchy, they would just make up a funny word or, you know, spell it backward or something," McElligott said. “Now, none of that’s going to work anymore."

The system is also able to decode colorful metaphors in communications, such as a trader suggesting that his colleagues “throw some chum in the water," a fishing reference that in a financial context can allude to illegal spoofing, McElligott said.

Some systems, like those offered by U.K.-based compliance provider SteelEye, can also automatically handle low-risk communications, freeing up compliance personnel to delve into trickier cases. That can be a selling point. Compliance isn’t a profit center, and often requires large numbers of staff focused on sometimes rote tasks, so labor-saving tools scaled over a large organization can save large amounts of money.

Listening up

A survey of staff at 250 financial firms released last month shows that only about a third of the firms use generative AI tools for any purpose. But 81% of firms with more than 500 employees said they have felt pressure to implement new tools, according to the survey, which was conducted by Smarsh, a communications data service provider.

Risk and compliance applications were the second-most commonly cited priority use case, after productivity uses such as summarizing meetings.

Regulators have urged a cautious approach toward AI. U.S. and European regulators have expressed concerns that an AI system might inadvertently be trained in ways that give it an improper bias, such as racial or gender.

Finra, Wall Street’s self-regulatory arm, said last year that it is considering issuing new guidance on the evolving uses of AI at firms. It said generative AI presents “promising opportunities" for compliance efficiencies.

Regulators also demand that institutions be able to explain the workings of systems that they rely upon, said Amy Jane Longo, a partner at the law firm Ropes & Gray. Companies using them also have to make sure they aren’t introducing a new source of cybersecurity threats.

“I’m both optimistic and cautious," Longo said.

Service providers rolling out the technology say they are looking to the next frontier.

Telephone calls are routinely recorded and retained on Wall Street. Firms haven’t been required to comprehensively monitor the thousands of hours of daily recordings an office might generate, but dramatic improvements in transcription technology now make such scrutiny a possibility.

“Historically, it’s been very expensive and complicated to transcribe voice, and the quality was not that good," said Matt Smith, chief executive of SteelEye.

SteelEye, which also provides AI to monitor financial transactions, offers a voice transcription service that it says can accurately render much industry jargon, and then feed the results into a system that can look for market abuses.

Smith said he thinks regulators could make automated snooping of employee phone calls a requirement.

“As regulators understand what’s possible now in the market with technology and the cost efficiencies that are now prevalent, I do expect in time regulators will become a lot more prescriptive," he said. “AI is an absolute game-changer for the financial markets."

Write to Richard Vanderford at Richard.Vanderford@wsj.com

Catch all the Technology News and Updates on Live Mint. Download The Mint News App to get Daily Market Updates & Live Business News.
more

topics

MINT SPECIALS