Ai Editorial: Outsmarting a Fraudster with Machine Learning

Outsmarting a fraudster with machine learning  

Machine learning automatically learns about new fraud patterns in real-time. Can it help in combating fraud? Ai’s Ritesh Gupta finds how it deals with fraudsters 

Travel brands are keenly looking at fighting fraud, revenue leakage and also curtailing associated costs.

In the era of omni-channel commerce, where airlines and OTAs need to embrace various forms of payment methods, companies face fraud on multiple fronts: on top of credit card fraud, merchants must deal with fraudulent accounts, abuse of promotional codes, and spammy content on their websites, like fake reviews or phishing messages. So how to keep a tab on such bad online behavior?

As a specialist, Jason Tan, CEO, Sift Science says machine learning is supremely suited to catching all of this.

“Think about how much customer data travel companies have access to: email addresses, billing and shipping addresses, phone numbers, device fingerprints. You also have behavioral data: the actions a user takes on your site, like where they click and what selections they make,” says Tan, who presented during Ai’s The Airline & Travel Payments & Fraud Summit, held recently in Fort Worth, Texas.


Machine learning can quickly and efficiently digest information to identify patterns, so you can start to tell a story about who your users are and what their intent is. When patterns of real-time fraud are mapped against examples of past fraud, merchants can accurately predict when they’re seeing a good shopper or a malicious one – so they can block the fraudsters, or make it easier for good customers to buy.

For example, as Tan says, Amazon uses machine learning to identify its good users and offer them 1-click checkout – a completely frictionless experience.  

Missing the bus

Tan categorically says if travel companies aren’t embracing machine learning for identifying the profile of fraudsters, then they're missing out on effective fraud prevention.

“Travel companies that resist implementing machine learning could instead be experiencing increased sales and better conversion rates by taking advantage of automation. You can use machine learning to create smart and dynamic checkout flows, where known good users can fly through purchasing, while additional friction points (in the form of cardholder verification) can be added for suspicious users,” he says.

Machine learning enables companies to automate aspects of fraud detection and make quicker decisions. Less time spent on manually reviewing orders means that companies reduce their overhead costs and can pass those savings along to their customers.

Consumers booking travel online expect their reservations to go through immediately. Travel companies don’t have the luxury of time; they need to automate parts of their fraud-detection process to stay competitive.

The team at Sift Science referred to several examples:

  • When TravelMob, a member of the HomeAway family of companies, started to notice credit card fraud on their site, their first approach was to manually review new booking requests. Not only was this time-consuming, but it was difficult to scale as their business grew. However, once they implemented a machine learning fraud solution, the TravelMob team became much more efficient. They automated aspects of the fraud prevention process, so the team ould focus only on the bookings that weretruly risky.
  • HotelTonight, on the other hand, experienced a 50% reduction in chargebacks after implementing a smart machine learning solution.
  • OpenTable reduced their manual reviews by over 80% and saved thousands of dollars in analyst man-hours by blocking fraudulent transactions automatically, before they lost any money.

Dealing with a fraudster

The most effective machine learning applications can take in and return information instantaneously, says Tan.

He adds, “For example, say I’m a fraudster that uses an email address like, and the business figures out that I’m bad because they get a chargeback. Through real-time offerings, jason123 is identified as a fraudster and the system immediately learns that people with 3 digits in their email address are more likely to be fraud. It doesn’t have to be jason123, it could be jason234, jason945, or fred579, but chances are good that those users are suspicious. So when I come back to that company’s site with or another fake email address, I would immediately be flagged as “probably a fraudster”.”

Sift Science’s “secret sauce” is its network of customers that send terabytes of data to its servers.

“That means all of our customers can benefit from the same learnings – for example, if we detect a fraudster on one site, that user’s Sift Score (a measure of riskiness) will instantly update across the entire network, so other businesses can block him. This feature enables companies of all sizes and of all locations not only get an individually tailored fraud prevention system, but also stay ahead of new and changing fraud patterns as their customer base grows,” explained Tan.

The data that companies choose to share should be based on their unique businesses and needs. There will be some common fields like departure destination that whole industries share, but there may also be company-specific data.

Global players like Airbnb and HotelTonight are able to use any data points that they already collect in order to benefit from machine learning for fraud. Details like stay length, airplane seat selection, and travel route can offer insights on top of more obvious ones gained from personal traveler information. A flexible machine learning system can take any data you throw at it.

As for visualizing fraud connections, Tan says a bad user might be testing hundreds of credit card numbers or have thousands of fake accounts on your site. “Using the data pulled from every order or transaction sent to Sift Science, we map out the suspicious signals that any given user or order shares with others.”

These connections help to identify why a user might be fraudulent, as well as allow merchants to proactively block users linked to past bad behavior.

Protecting data

It is imperative for travel companies to ensure that attacks don’t affect credit card data as well as any other personal passenger data.              

Unfortunately, it’s getting harder and harder for companies to “ensure” that data stays secure.

Data breaches will soon be the new normal, says Tan.

“Although machine learning can’t stop hackers (yet), it can help travel companies ensure that stolen data isn’t successfully used on their sites. Employing a machine learning solution can actively identify suspicious behavior, and prevent a chargeback for the merchant, and a painful fraudulent purchase for the victim,” said Tan, answering a vital question.

Real-time aspect

One of the best things about using machine learning is that it automatically learns about new fraud patterns in real time so you don’t have to keep close tabs on new tactics. Travel brands rely heavily on online transactions, so there is also a need to watch out for new travellers from new locales. Travel brands need to be mindful that new geographies come with different types of fraudsters and fraud patterns. A pattern may be normal in one region but fraudulent in another, said Tan. Of course, you can’t just block every new traveler; that would be a quick way to lose legitimate business. But leveraging big data to weed out the bad users wielding stolen credit card numbers is key.