The use of machine learning in fighting fraud

How does pattern recognition and machine intelligence help retailers, financial institutions, and consumers fight online fraud?

Machine learning is truly revolutionizing fraud detection by giving online businesses of all types actionable insights about their site visitors. Think about the amount of data that online businesses have access to: everything from email addresses, phone numbers, and device fingerprints to behavioral data like where people are clicking and how much time they’re spending on different pages. Machine learning can quickly and efficiently digest all of that information and find patterns in real time, delivering an accurate picture of an individual user’s intent.

Is someone on your site for a legitimate reason, or a malicious one? Once you have an answer to that question, you can make the best decisions about what type of experience to deliver – making it as simple as possible for good users to sign up and check out, blocking fraudsters, and asking for additional authentication information if needed.

Machine learning also provides significant operational savings, enabling businesses to automate a large portion of their fraud management. Online payment provider Entropay used Sift Science’s machine learning solution to grow conversions by 15%, while shrinking the time spent focused on fraud by 90%.

Which are the advantages and disadvantages of a rule-based fraud detection approach?

I understand why some merchants are reluctant to move away from rules – you can apply them in a targeted way, which gives you a certain measure of control. And some businesses have hard-and-fast business rules that can restrict dealings to certain geographies.

However, there are significant problems with relying on a rules-based approach to fighting fraud. First, rules don’t adapt and don’t learn. That means they’re bad at predicting new fraud attacks and keeping up with today’s agile and technologically advanced fraudsters. You leave your business vulnerable.

Second, rules are notorious for producing false positives and blocking good customers. We’ve seen businesses significantly increase their conversion rates – specifically in new markets – when they stopped relying primarily on rules.

Third, rules require significant resources to maintain and update. Scaling with rules becomes quite a burden for fast-growing companies, who may resort to adding headcount to fraud teams to combat these limitations. Our customers who move from rules to machine learning are amazed at how little time they need to spend on tedious tasks, and how they no longer need to devote data science resources to fraud or build out manual review teams.

As consumers are getting more used to a seamless customer journey for on-demand services and digital goods, how should a merchant address the challenge of recognizing good customers in real time?

These days, all online businesses are competing on speed and convenience. Big players like Amazon and Uber, with their one-click checkouts and streamlined signup, have set a high bar for what consumers expect when doing business online.

While physical goods merchants may have hours or even days to fulfill an order, on-demand and digital goods merchants have minutes – or less. This means many of our customers who use these business models don’t even manually review new orders. Their business depends on a highly accurate, real-time technology that they can use to automatically accept or reject new orders based on risk. Machine learning is the only technology that can really stand up to this challenge.

Which are the most popular schemes and techniques used by criminals to create false accounts and how can merchants and financial services protect themselves against them?

Fake accounts are low-hanging fruit, providing an easy foot in the door for fraudsters and scammers to carry out all kinds of malicious behavior – from spamming to phishing and identity theft. We see bad actors creating accounts en masse using disposable email domains and email addresses, where only a single letter or number has changed. Technology makes this even easier, as fraudsters can use software to create hundreds of new accounts in minutes, or automate the creation of fake social media profiles by using private proxies, randomized page load times, and scraped data.

There are a number of tactics you can use to proactively prevent phony signups, like adding a Captcha or SMS verification step. However, fraudsters have ways of bypassing even those checks, and friction of any kind can turn off legitimate users. That is why machine learning is an ideal solution for identifying risky users in the background, without any added friction for low-risk users. You can automatically block bad actors from signing up on your site, saving time, money – and your business’ reputation.

This interview is also featured in our Web Fraud Prevention and Online Authentication Market Guide 2016/2017, a must-have resource for up-to-date information on the ever changing world of web fraud and online authentication.

About Jason Tan

Jason Tan is the Co-Founder and CEO of Sift Science. He previously served as CTO of machine learning startup BuzzLabs (acquired by InterActiveCorp), and as an early engineer at Zillow and Optify.

About Sift Science

Sift Science is a machine learning company focused on establishing trust between businesses and consumers by optimizing user experience and preventing fraud. The Sift Science Trust Platform hosts a full suite of products attacking every vector of online fraud and abuse: account takeover, payment fraud, content and promo abuse and fake accounts.