Some rules are black and white, which is why Sift Science has added the ability for clients to integrate automatic rules on top of its machine-learning fraud detection system.
"This was informed by our customers who say, 'We have this hard and fast business policy,'" said Jason Tan, CEO of Sift Science. "There are situations where you need a rules-based system because you know for a fact you want to take action against a certain criteria."
For example, a particular U.S. merchant might never want to ship products to North Korea, Tan said.
Sift Science is also allowing merchants access to a customer's device ID through an application programming interface (API), giving merchants more insight into device fingerprinting. The company has used device fingerprinting since its launch about four years ago, but hadn't allowed merchants to control that feature.
Since the beginning, Sift Science has focused on its machine-learning software, which is customizable by vertical and geography. The company also pools data from across all of its merchant clients in real time to make fraud decisions.
"Machine learning can predict and now with rules-based and device fingerprinting you can enforce and act," Tan said.
While merchants are asking for rules-based systems, machine-learning fraud detection is not losing momentum. It's become a popular technology for startups, including Feedzai and Riskified, to innovate with.
And as mobile proliferates, device fingerprinting won't be a signal merchants can rely on anymore.
"My iPhone and your iPhone will look the same," Tan said. "So merchants will have to start looking at customer-specific behaviors," such as how fast a customer types their username and password.
Sift Science says its technology improves both fraud detection and the consumer experience. "Lots of merchants oftentimes treat their customers like airport security," Tan said. In both scenarios, most people aren't acting maliciously but "the majority ends up paying for the actions of the minority. And that's a bad experience."
Sift Science's platform reduces the false positive rate from an industry average of 80% to about 20%, Tan said. Lowering this rate cuts costs by reducing the need for merchants' reliance on fraud analysts for these transactions.
Sift Science has grown its client base by five times since last year, as several different industries become more familiar with machine learning software. Sift Science also serves clients in the remittance, digital wallet, digital gift cards and digital currencies markets. These merchants aren't just seeing fraud from chargebacks, Tan said, but they're also catching fake account, account takeover and referral fraud.