In 2006, after obtaining his doctorate at Carnegie Mellon, Yinglian Xie was not thinking about whether he wanted to make a career in the United States or his native China, much less to one day becoming a businesswoman. Instead, he was very focused on the best place to continue his work: his thesis was about the identification of possible threats to internet security by searching for correlations between apparently unrelated events. It ended in the Laboratories of Silicon Valley of Microsoft Research, a research center on technological advances. “They were really the best in the world,” Xie was marveling, noting that among his colleagues there were several winners of the Turing Prize, also known as the Nobel Computer Prize.
Three weeks later, after obtaining his own doctorate at the University of California/Berkeley (also with a thesis on Internet security), Fang Yu reached the same laboratory. The women had grown 30 minutes away from the city of Suzhou, west of Shanghai, and quickly became friends and scientific collaborators.
Today, Xie, 48, and Yu, 46, have dozens of academic articles and 11 years of struggles with startups behind them. Now, the company that co -founded in 2013, Datavisor, based in Mountain View, California, has finally managed to make their way in the protection area of financial firms and their clients against fraud. With Xie as CEO and Yu as product director, Datavisor’s revenues fired 67% in 2024 to 50 million dollars, which helped him appear for the first time on the Fintech 50 list forbes which recognizes the most innovative financial technology startups in the United States. Its clients include Sofi, AFFIRM and Marqeta.
At present, banks and financial technology companies use multiple techniques and security suppliers in a true arms race with scammers; Datavisor’s niche is to find emerging fraud networks before they can inflict great losses. These networks find new ways to seize user bead groups, exploit vulnerabilities such as credit information leaks to send fraudulent loan requests or deceive off guard to pay for illegitimate products. Every time a new network or fraud method is discovered, fraud detection models can be updated to detect those alert signals, but that does not always help users who have already been affected.
“With typical automatic learning, it is necessary to train it to learn and improve,” says Xie. “It is always reactive and attack patterns are detected months ago, when things have already moved and changed.”
Datavisor’s secret ingredient is what is known as “non -supervised” automatic learning, which uses algorithms to analyze data sets that are not labeled and discover correlations by themselves without humans telling him what objectives or categories to look for. Xie will not go into details about what makes patented algorithms work, but gives a realistic example: let’s imagine that a fraud network obtains access to a bank’s data and identifies a certain victim profile, as old clients who have amounts of High average transaction and less digital experience. Then they would be sent fraudulent offers to buy a gift card and, if the dollar amount is below their typical transaction, it is less likely to activate the bank’s fraud filters. But non -supervised automatic learning of datavisor could make connections between these bank customers in milliseconds (connections that nobody told him to look for) and block the avalanche of false offers in real time.
“What makes Datavisor unique is that we can do groups in real time,” says Yu. “Every day or even every hour new schemes arise.” That capacity is particularly valuable these days, says Xie, because “practically all important attacks today come from these coordinated fraud networks.”
In fact, a recent report from the AU10IX identity verification company declared that 2024 would be the year of “fraud as a service”. The average number of incidents in each coordinated “megaatque” doubled from 4,000 to 8,000, says the report, and this commercialization of crime is increasingly removing consumers. Fraud losses reported to the Federal Commerce Commission in the US.
This is valid for Xie and Yu: they have the necessary technical knowledge to support their claims that their algorithms are superior. Both were university superstars: Xie held first place among 140 computer students who graduated with her at the University of Beijing and Yu conducted practices with the founding members of Microsoft Research Asia while studying at the University of Fudan in Shanghai, which the inspired to do a doctorate.
Both arrived in the United States to carry out postgraduate studies, believing that it was the ideal place to study avant -garde computer science. They obtained permanent residence and stayed to work for Microsoft, where they finally became citizens.
In the seven years who worked in Microsoft, both published dozens of articles that have since been summoned thousands of times. They often collaborated as co -authors in published articles that covered topics such as a new approach to detect search robot trafficking or how to identify malicious web advertising schemes.
“We had many ideas, but we were always waiting for other people to retake them to make them a reality. We talked about if we stayed in Microsoft Research for another year, we could publish three or four articles a year, but after many years, one disappoints with that level of impact, ”says Xie. “We wanted something more real.”
They had also contacted them researchers from other companies such as Yelp, Pinterest and Facebook who had read their articles and wanted to collaborate in similar problems of data analysis. So in 2013 they made the leap to the business world.
“Before founding the company, we asked people if we were prepared,” says Yu. “The unanimous answer was: ‘No, you don’t know what awaits you.”
Starting with their own savings and connections in Silicon Valley, they got a couple of initial clients such as Yelp, which wanted to identify if users were abusing their system with reviews, and the Chinese application of Momo instant messaging. They closed a series A of 14.5 million dollars in 2015, the first external money they achieved, and created a niche in the creation of security solutions for high -tech Internet companies. In 2018, Datavisor raised another 40 million dollars led by China Sequoia, and reached an assessment of 390 million dollars according to Pitchbook after 12 million more in 2019.
But, deep down, his market was running out. While datavisor focused on abusing the promotions that technology companies offered to attract more users, their customers did not offer so many of these rewards, which normally did not generate high retention rates. At the same time, financial institutions were digitizing their systems much faster and Datavisor changed their strategy to address them as clients.
The change was not fast. Xie says that two or three years were needed to “turn our product” and make their algorithms more integral, such as a unique window for fraud prevention. Datavisor raised $ 40 million in December 2022, led by Brighton Park capital, to help in his reinvention. According to Pitchbook, he needed a lower assessment of 260 million dollars for that round, at a time when investors’ enthusiasm for the Fintechs was declining, which raised its total financing to more than 100 million dollars. Forbes He estimates that the two founders have collectively preserved around 25% of the company.
Xie describes the change in the feeling of the market in 2022 as a “review of reality” for her and her board of directors after the previous year established a high bar for the assessments. But he never felt serious pressure to bring more experienced managers to raise another round. On the other hand, he says that Datavisor ended with multiple leaves of conditions to consider and chose to associate with Brighton Park, a growth capital firm based in Greenwich, Connecticut, which he believed in his long -term vision.
“It was a time when we needed additional financing to move on to the next level,” says Xie. “We were very sure that we still had the best technology in the world.”
Until now, the new image has been successful. Among the 50 Datavisor customers are AFFIRM, a company that offers immediate purchase and subsequent payment services, the Sofi Digital Bank and the Marqeta card issuer. Although this amount of customers is insignificant compared to identity verification companies as a person or socure, who provide services to thousands of companies, Datavisor promises a deep relationship that covers everything, from the incorporation of users to the monitoring of their transactions and bank transfers. Customers pay an annual subscription rate that varies according to the volume of events that need Datavisor to process.
Xie states that many clients go to Datavisor in search of consolidating their efforts to fight fraud after discovering that trying to integrate several different suppliers generates headaches and inconsistencies. This comprehensive and accompaniment approach allows datavisor to charge more for each additional client than the competition. “Over time, we hope that we open a way to become a much larger company,” he says.
This article was originally published by Forbes Us.
You may be interested: know the first ‘Chili -onaire’ of Mexico: Horacio Fernández, founder and CEO of Tajín
