Tim Estes, founder of Digital Reasoning, discovered his philosopher’s stone using algorithms to turn typo-ridden emails into gold
Image: Evan Kafka for Forbes
Talk is cheap, unless you’re one of the biggest banks in the world. Then, offhand, typo-filled instant messages and emails among traders, such as “dont want other numpty’s in mkt to know” or “hopefulyl a fe w more get same way and we can team whack it” or “i’d prefer we join forces... lets double team them,” can lead to billions in fines and even criminal charges.
Since the financial crisis, regulators have seized on traders’ emails and Bloomberg instant messages to establish collusion and market manipulation in the foreign exchange, commodity and interest-rate markets. JPMorgan, Citigroup, UBS and Barclays have all pleaded guilty to felony charges, and just about every big bank has paid fines exceeding $100 million. Today, detecting conspiracies in traders’ cryptic communications—before government enforcers do—is a critical part of risk management.
Surprisingly, one of the banks’ best weapons comes from a glass-and-steel office park 20 miles outside Nashville. That’s the headquarters of Digital Reasoning, a 17-year-old machine-learning pioneer whose Synthesys technology for surveillance of natural-language communications is used by Nasdaq, Goldman Sachs, UBS and billionaire Steven A Cohen’s Point72 Asset Management as well as by US intelligence analysts tracking terrorists and 1,000 law-enforcement agencies involved in the fight against human trafficking.
On a quiet Friday afternoon at the headquarters, Digital Reasoning’s beer tap looks out of use and weekend talk among employees surrounds church activities and a Nashville Symphony tribute to David Bowie. Founder Tim Estes, a 37-year-old Tennessee native who sings in the choir at his Presbyterian church, sets the tone. Dressed in a professorial grey cardigan and ill-fitting dark blue jeans, he walks to a whiteboard and writes out a sentence in a purple dry erase pen, then starts tagging significant words and the distance between them with binary numbers.
“Meaning is indistinguishable from use when it comes to language,” he lectures. “The meaning of a word is really driven by the expectation created by the context around the word over time.” Once Estes realised this in the early 2000s, he started creating a patented set of algorithms that could approximate and expand on the sort of context humans apply to what they read.
Synthesys works in ten languages, linking people, places and things, similar terms, aliases and semantic labels and even incorporating how word relationships change over time. For each industry, algorithms get special training. For the banks, along with the financial lingo, Digital Reasoning will “deeply train” its algorithms on certain word patterns that could set off a compliance alert, Estes says. (Hint: Calling suckers “numptys” in an email to your co-conspirator might give you away.)
The biggest tell in Wall Street messaging? Boasting. “People often try to conceal. What they can’t help themselves from doing is to congratulate themselves afterwards,’’ says Valerie Bannert-Thurner, head of risk and surveillance solutions at Nasdaq, which is in the process of integrating Synthesys into its trading surveillance system. “Nobody, from what we have seen, has the natural-language processing skills, understanding and deep learning that Digital Reasoning has,’’ she says. “It is a quantum leap.” Nasdaq isn’t exactly a disinterested party; it invested in Digital Reasoning’s last funding round in May 2016, which took in $40 million and valued the company at more than $200 million. Estes remains the largest shareholder.
Neither flagging fraudsters nor getting rich was Estes’s goal when he first began developing his technology. He was a University of Virginia undergraduate who had enrolled to study engineering before switching to philosophy, filling his spare time singing in the glee club and Black Voices, a gospel choir. Given his eclectic interests, Estes thought it would be useful to have an online system that would recommend events on the busy campus based on a user’s networks and activities. The student council declined to fund his proposal, but Estes discovered his intellectual passion: Using math to discover meaning and connections in language. “It was about looking for algorithms that could tease the order out,’’ says Estes, adding that he has “a strong philosophical and theological bias” that the world is designed and has order, even if we don’t always see it.
In 2000, in his senior year, Estes launched Digital Reasoning with Kenneth Elzinga, now chair of UVA’s economics department, as his biggest financial backer. He had less than $1 million in startup cash and figured he would need a deep-pocketed customer to develop a working system. He started talks with US army intelligence shortly before the 9/11 attacks and by the mid-2000s had built a system, called Interceptor, that enables analysts to search intercepted messages for links to terrorism.
Still, as Estes discovered the hard way, living off government contracts, which originally accounted for 100 percent of his business, can be hazardous for a small company. In 2008, Digital Reasoning had a near-death experience: The expiration and renegotiation of a federal contract that had provided the bulk of its revenue led to a half-year gap in payments.
Estes couldn’t meet payroll for his 15 employees and took a second mortgage on his house to cover health insurance premiums for workers. Some employees worked for months without regular paychecks. (Their dedication paid off; they got equity and, later, bonuses.) In 2009, Digital Reasoning’s revenues exceeded $5 million, up from $1.5 million in 2008. In late 2010, the CIA’s venture capital arm, In-Q-Tel, invested in Digital Reasoning. (Estes won’t say how much.) With the company now healthy, Estes turned to his next market: Financial services. “There is a long-standing tradition in analytics and security of banks adopting technology from the intelligence community,” he says.
Nasdaq invested in Digital Reasoning’s last funding round which valued it at over $200 million