It is no secret to e-commerce merchants that Amazon, one of the giants in its field, has succeeded in part because it can make fast decisions about whether a transaction is legit or has fraud written all over it. And these decisions have to look at more than just whether a shopper is using a stolen card.
For the past five years, Sift Science has been helping other online retail, financial services, travel and ticketing marketplaces gain that same expertise through machine-learning fraud protection.
Investors have banked on San Francisco-based Sift Science in the past as a company that can deliver high-level transaction protection and fraud prevention tools to an industry facing increasing cyber threats. That confidence continued this week as the company enjoyed a $30 million series C funding round, while also revealing a new set of products to detect more of the fraud plaguing e-commerce.
"When we started the company, we focused on payment card fraud and reducing chargebacks," said Sift Science CEO Jason Tan. "Over the last few years, in working with our customers, more than 6,000 sites, we learned that fraud comes in all shapes and sizes."
Similar client feedback in the past helped the company determine that it needed a more powerful analytics tool, because so many e-commerce merchants were using rules-based systems that let too many unknown fraud techniques slip through.
In addressing feedback and advancing cybercrime threats, Sift Science is adding account, content and promotional abuse as areas of detection, as well a fraud prediction tool to create specific risk prevention strategies. The new abuse prevention measures will help identify fake account registrants; pinpoint fraudulent listings, messages and inquiries; and target fraudsters who manipulate online coupons and rewards programs.
"Many merchants suffer from all types of fraud, and they want to fight that with one platform, rather than through different pieces," Tan said. "We want to be that scalable platform."
Sift Science feels it is in a good position to be the platform, saying it already helps customers deny more than 9 million fraudulent charges each year, saving those merchants more than $1 billion. The company estimates, on average, that it saves merchants $140 per fraudulent transaction prevented.
"Machine-based learning is really strong, because roughly speaking only about half of threats, though some say maybe 80%, can be detected using rules," said Avivah Litan, a vice president and fraud analyst at Gartner Inc., a Stamford, Conn.-based market research company. "So that leaves another 20% of transactions that you don't know what to ask, because you haven't seen it before, or the criminals know how to beat the rules because they always do a job of figuring that out."
While rules-based prevention systems focus on parameters such as transaction limits over a certain time period, or blocking certain geographic regions, machine-learning represents a far deeper dive into data analysis.
"I would call it advanced analytics that includes machine learning," Litan said of the technique. "Humans just can't go through all of this information the way a machine does because the machines have models for deep learning."
That deep learning can detect, for example, how fast a certain consumer or cardholder generally types in an online order or navigates a site. But mostly, the "machine" screening the transactions knows what it needs to look at, what is important and what anomalies appear in the data sets, Litan said.
"You could never do that with the human brain," she added. "You really need advanced analytics to stop the really sophisticated fraudsters."
As one of the few companies — along with NuData Security and Distil Networks — offering machine-learning protection, Sift Science is trying to make it easy for a merchant to integrate the system.
Sift Science offers a dashboard portal for merchants to monitor the cloud-based software-as-a-service, showing why certain transactions or site visitors were scored a certain way, with some visual and graphic tools to aid the process.
"For most customers, it is all about automation, being able to decide in the moment whether this user can be trusted or not," Tan said. "Status quo solutions can minimize fraud, but they do so with a compromise because they force merchants to staff large manual review teams to look over transactions."
The dashboard illustrates what Sift Science is doing to protect a site, but "the end game is accuracy" for most merchants, Tan said. "Merchants that want to operate at massive scale, but have to automate in real time, appreciate our speed and accuracy," he added.
The company received its recent financial support from Insight Venture Partners, with contributions from Union Square Ventures and Spark Capital.
Sift Science, serving clients such as OpenTable, Hotel Tonight, Instacart and Zoosk, says it currently has a total funding figure of $54 million.