By Sophia Stuart

Online fraud is a perpetually growing problem for retailers, financial institutions, and consumers in general, but Sift Science believes it has the solution, thanks to pattern recognition and machine intelligence.

Incubated at Y Combinator, Sift Science launched in 2011. CEO Jason Tan was just 25, but already had experience at Seattle start-ups such as Zillow, Optify, and Buzzlabs, where he was CTO when it was acquired by IAC.

“We wanted to find a problem to solve and we knew machine learning was the future so we wanted to apply that,” Tan told PCMag about the origins of Sift. “We asked friends at other tech companies about the problems they faced [but] didn’t have the capability or desire to fix in-house. Fraud was the word we heard over and over again. So we dug deeper and realized that we’d found our opportunity: to apply best-in-class machine intelligence to fraud detection.”

The company today includes experts in speech recognition, digital payment systems, e-commerce analytics, sentiment analysis, and text recognition, who keep an eye on consumer behavior for clients including Yelp, Hotel Tonight, and OpenTable.

Here’s how it works: Sift Science takes over 5,000 identity and behavioral signals and does a real-time risk assessment on each user as they attempt to complete a transaction. It’s constantly learning and drawing on its global network to update data on the fly, partly through Java snippets on clients’ sites, as well as rendering and triangulating profile information on identical IP addresses recorded throughout its ecosystem. The machine intelligence is constantly churning, analyzing, predicting, and issuing warnings, or approvals.

It also embeds unique data points from different sectors. For instance, in the shoe business, a customer who generally orders a size 12 and then suddenly switches to a size 9 might be an indication of being up to no good. Conversely, consumers in the hospitality arena who are known to block book rooms and then cancel at the last minute, causing headaches for hoteliers, can be flagged by Sift Science. The mind boggles at the distinct interpersonal rules Sift Science inserted to manage consumer behavior for client Zoosk.com.

“We claim 10X results above other solutions,” Tan said. “And the ultimate benefit of machine learning is a probability engine that goes both ways. We’re not just looking for motivated and sophisticated human adversaries in the escalating fraud problem, we’re also verifying good customers, trustworthy people, and helping clients dynamically adjust behavior to reward this, like reducing friction (less form fields or CAPTCHA) in the checkout process in real time.”

So should customers care that there’s a raft of Sift Science servers keeping a beady eye on their shopping excursions? Tan said he often gets a “so what?” shrug from people when he tells them what he does.

“The most common response is ‘Oh, yeah, fraud happened on my account but my credit card dealt with it,’ but they should care,” Tan pointed out, “Because brands that they love depend on keeping fraud at bay. Many beloved brands become a victim of their own success. It’s only when they reach a certain level of profitability and revenue size that the fraudsters attack them.”

In the fight against crime, Sift Science is silently keeping watch to ensure its merchant clients can stay in business. So next time you’re engaging in digital commerce, just remember, your browser is judging you.