Recent major data breaches at Sonic, Hyatt, Whole Foods – and, of course, Equifax – have put all merchants and financial institutions on high alert. With compromised information in the hands of the wrong people, your business is at risk. Merchants spend 7% of their annual revenue on fraud every year – including the cost of chargebacks, fraud management operations, and false positives – according to Javelin Strategy and Research.
At the same time as fraud is increasing, consumer expectations for digital interactions also continue to rise. These days, all online businesses are competing on speed and convenience. Businesses that wrongly block trusted users or provide too many hoops to jump through at checkout will lose out. How do you balance the need to protect your business and users with your goals to maximize conversion?
Legacy fraud detection systems can’t keep up
Traditionally, companies have implemented rules engines to combat fraud and abuse. However, there are a number of ways that rules come up short. First, they just aren’t accurate enough. They treat the world in black and white, leading to both false positives (users wrongly blocked as fraudulent) and false negatives (actual fraud that is missed).
Second, rules don’t adapt. Fraudsters can easily figure out what rules are being used and change their approach. But by the time a new fraud pattern has been identified, bad orders have already gone through.
Third, rules-based systems require constant evaluation and updates. You end up with hundreds of rules that you can neither keep track of nor easily scale back if they are ineffective. And they require a large investment in people who can both update the rules and keep up with orders that are flagged for manual review.
Machine learning: a more scalable approach
So, what are the alternatives to rules? One approach that is often used in place of rules, or in conjunction with them, is machine learning. Machine learning collects passive data about users’ behavior on a site, like where they’re clicking and how much time they spend on different pages, and combines it with identity information such as email addresses, phone numbers, and device fingerprints.
A machine learning system can quickly and efficiently digest all of that information in real time, and identify patterns that indicate whether a user has fraudulent intentions. Then, you can make sure you’re delivering a seamless experience to your trusted users, blocking fraudsters from exploiting your site, and asking for additional authentication for gray-area cases.
Not only can machine learning systems deliver real-time decisions, but they continuously learn from all of the quality data and feedback flowing back into the system, growing increasingly accurate in their predictions. The upshot is that you can accurately identify and stop fraud before the transaction goes through, without compromising user experience.
In fact, with a machine learning system you can also proactively identify your good users and provide them with an streamlined payment experience – by including fewer form fields and identity checks – that increases conversion. Fraud may be a formidable problem, but fortunately merchants have access advanced technology that can both protect them and power their growth.
About Kevin Lee
Kevin Lee, Trust and Safety Architect at Sift Science, builds high performing teams and systems to combat malicious behavior. He has lead various risk, chargeback, collections, spam and trust and safety initiatives for organizations including Facebook, Square and Google.
About Sift Science
Sift Science Live Machine Learning, global trust network, and automation technologies fuel business growth and power expansion into new markets while protecting businesses and their customers from all vectors of fraud and abuse.