Modernizing Authentication — What It Takes to Transform Secure Access
When Jason Tan and Brandon Ballinger decided to launch a startup, they knew two things: They wanted to leverage their team's impressive engineering backgrounds, and they wanted to build something with machine learning. All they needed was to find the right problem to solve.
"In that exploration process, we talked to a bunch of our friends who worked at different companies and asked them, 'Hey what are problems that your business faces today that your business would rather have someone else solve for you?'" said Tan, CEO of Sift Science. "And fraud often came up as a problem that was a thorn in the side that none of these businesses wanted to solve themselves. It's something that needed to be solved but they didn't want to make it a core competency to figure out how to be world-class at fraud."
That feedback indicated a strong opportunity to build a business, Tan said. Yet neither engineer had a background in fraud or risk or even payment.
"When we started the company, we literally did not know what a chargeback was," Tan said with a laugh.
Their research, however, found the space was ripe for innovation.
Machine Learning Meets Fraud Prevention
"It's been stuck in the dinosaur ages with these rules-based systems, and the improvement every year from these existing vendors is more like 10 percent every year rather than 10 times," he said. "We now have the technology with machine learning to do 10 times better. The analogy I like to use is email spam filters. Twenty years ago, email spam filters were all primarily rules-based systems and they weren't that effective. But today if you use Gmail or any kind of decently good email system, its spam filter is primarily powered by machine learning and it's all fully automated."
Tan and Ballinger had found their problem. So in 2011, with a small team of engineers with credentials including Facebook, Google, Amazon and Zillow, they launched Sift Science.
"We're Sift Science because we sift through Big Data using computer science, plus alliteration is always fun," Tan explained.
While machine learning is more commonplace now, at the time it was on the cutting edge of fraud prevention. Tan recalled the first Merchant Council Risk Conference he attended in 2013. After seeing a sign in the Sift Science booth that read "Fight fraud with machine learning," an executive from a Fortune 1000 company asked Tan, "What's machine learning?"
While Tan and Ballinger had the technology, they still had to show businesses how it could benefit them.
"That was a big wake up call," he said. "What we're trying to bring to the market is so cutting edge compared to what people are used to, so our work is cut out for us to educate and evangelize the benefits of machine learning systems."
The company has an impressive, and varied, list of clients. While fraud detection technology vendors typically focus on financial services, Sift Science's client list includes online giants such as Zillow, Twitter, Yelp, Indeed and Match.com. The company doesn't position itself as simply solving fraud.
"We are in the business of helping businesses decide who they can trust" Tan said. "Status quo solutions have traditionally been rules-based systems that are very reactive and unscalable and inefficient and basically inaccurate."
Using machine learning allowed Sift Science to shift from preventing risk in a wholesale way to a more targeted approach. Tan compared traditional rules-based tools to airport security post 9/11; the assumption is that everyone is bad.
"If you think about airport security … Most of us are not terrorists or have malicious intent, but we are treated as if we potentially are. We have to take our shoes off and do all this stuff unnecessarily," he said. "How do we instead have a more intelligent system that is able to provide a TSA-free type of experience for the deserving population and weed out the truly suspicious people and do so based on data?"
A machine learning system learns dynamically what bad behavior looks like instead of relying on pre-programmed rules. In effect, it learns from the data. This also reduces false positives and false negatives, Tan said, noting that his company's false positive rate is usually less than one percent, compared to the traditional software industry rates of 10 percent or more.
More than Fraud Prevention
More significantly, the underlying machine learning means Sift Science offers more than fraud prevention. It can also make online experiences less annoying for customers, and that can lower cart abandonment -- a significant problem for online businesses.
For instance, a rules-based system would require every customer to type in a captcha or include the three-digit security code on the back of a credit card to finish the transaction. Sift Science is able to eliminate that requirement except in cases that present greater risk. For the vast majority of customers, this removes a barrier that may have resulted in a lost sale.
"The more easy you can make it for them to complete the order, the less cart abandonment you're going to see," Tan said. "So you're going to start capturing a small but meaningful percentage of customers that previously would've abandoned the cart but now aren't."
The startup also claims to save clients an average of $140 per fraudulent transaction prevention. Add the numbers up, and it's not hard to see how Sift Science has expanded beyond the traditional financial services and retail industries into other verticals.
The cost to customers is based on transactions, which means Sift Science's charges align with their clients' growth. Some customers pay as little as $10 a month while others pay as much as $40,000 a month or more, Tan said.
Sift Science's platform is already leveraged for payment fraud, account abuse, promo abuse and content abuse. Tan foresees more growth in fraud use cases such as account takeover or targeting "friendly fraud," where buyers refute the charges to credit card companies as a way of avoiding payment. The company also is expanding internationally, and is looking at ways to make proactive investments to support that growth.
Fast Facts about Sift Science
Founders: Jason Tan, Brandon Ballinger
HQ: San Francisco
Product: Sift Science provides real-time machine learning solutions that help online businesses prevent fraud and abuse.
Customers: AirBnB, OpenTable, Instacart, HotelTonight, Wayfair, Kickstarter, Twitter, Match.com
Funding: $53.6 million in four rounds
Loraine Lawson is a freelance writer specializing in technology and business issues, including integration, healthcare IT, cloud and Big Data.