Flying High Again with a Better Online Travel Purchase Experience

Travel booking websites were hit hard during the worldwide lockdowns during the global COVID-19 pandemic. Steep drops in airline travel meant every booking was critical, and delighting customers was essential to ensuring loyalty and repeat business. Furthermore, data privacy and security is more important than ever, as any breech becomes front page news, potentially damaging the brand. As travel volumes begin recovering in certain regions, these businesses find that the competition is intensifying as new entrants come online. However, one of the highest priorities is discerning a value-conscious traveler seeking an affordable airline ticket from fraudsters seeking to buy tickets with stolen credit cards.

I sat down with Drayton Williams, Fraud Investigations Manager of Flighthub and justfly.com, an online travel company, to discuss their decision to use Amazon Fraud Detector and how it positions their business to grow as consumers return to the skies.

AWS: Who is FlightHub?

Drayton Williams: We think of ourselves as a tech company specializing in travel rather than a travel agency with websites. We make travel accessible, allowing more people to visit new places and explore different cultures by offering the cheapest flights to worldwide destinations. This is achieved by investing heavily in both technology and people, as well as by streamlining and optimizing the customer experience.

AWS: What is your fraud challenge?

Drayton Williams: We have tight profit margins, so we must balance the need to detect and stop fraudulent checkouts with ensuring that we’re not rejecting checkouts from good customers. Every checkout resulting in a chargeback means we both lose profit on the sale, and have to “eat the cost” of the tickets sold. Therefore, spotting the checkouts most likely to result in a chargeback is essential. Simultaneously, we can’t solve our fraud problem through blocking or increasing friction on legitimate potential business. Maintaining a low churn rate – the rate at which potentially good orders are blocked or abandon a purchase – is critical for us to capture all meaningful business from good customers. Of course, doing all of this while fulfilling customer orders as rapidly as possible is also essential to providing the great customer experience that results in repeat business, and allows our business to thrive.

AWS: Why did FlightHub choose Amazon Fraud Detector?

Drayton Williams: Our original fraud detection solution couldn’t match the unique challenges of a travel business. It was rules-based, and we found that adapting it to our needs was difficult. So we pivoted to building our own fraud model. This worked well until COVID hit, with “normal” travel patterns shifting from people visiting friends and family or going on vacations to people seeking to relocate due to the pandemic. That change made it difficult for our model to accurately identify and distinguish the new patterns of legitimate activity from fraud, which prompted the exploration of other options.

I did some digging and found that AWS offered a fraud detection service utilizing machine learning, which would help us by identifying our new fraud patterns faster. It also promised ease of use and accurate models customized to our business, all without a huge amount of data. So I was encouraged to try it.

I started with historical checkout data including 20-25 attributes, ranging from account information, purchase details, how many previous bookings the user had, and how long ago the purchases occurred. I happily found how easy and self-explanatory using the service was – training and deploying a machine learning model to detect checkout fraud from our data. I was also pleased with the ease of re-training a new model version from an updated data set, which helps detect new issues that we identify.

Based on the results obtained, and the convenience of use, we launched Amazon Fraud Detector into production in February 2021 to evaluate suspicious checkouts flagged by our model, and to automatically determine which of these checkouts should be allowed to process and which should be blocked. It’s supported our business ever since, as we experience a surge in bookings with consumers beginning to travel again.

AWS: What results have you seen so far from using Amazon Fraud Detector?

Drayton Williams: One of the first benefits we noted immediately was the impact of a fraud ring that attacked us after launch. Our old system had difficulty recognizing and adapting to the new attack pattern. With Amazon Fraud Detector, we found that a newly trained model could detect the new fraudulent transactions instantly, thereby limiting the impact of the attack.

Since introducing Amazon Fraud Detector, our abort rate has dropped below 2% (vs. 5% previously). Additionally, our chargeback rate is the lowest it’s ever been since the company’s inception. The business can now accept more checkouts that our past models would have flagged as risky and turned away. But perhaps the best thing is we’re getting these great results with roughly the same operational costs as before. All of this results in an increased number of bookings and revenues, along with a decrease in losses due to chargebacks.

AWS: What future plans do you have to expand your use of Amazon Fraud Detector?

Drayton Williams: We’re very satisfied with the service, and expect to use it even more as demand returns, the industry recovers, and we handle more booking requests. In addition, we’re developing version 2 of our current fraud model integration to introduce more input attributes. We expect to leverage the new Amazon Fraud Detector features and optimize our transaction fraud detection. And we anticipate this will provide a significant boost to our existing detection.

AWS: Thank you for chatting with us, Drayton. We appreciate your insights and expertise.

If you have questions or other feedback for Drayton, FlightHub, or AWS, leave it in the Comments section. For more information about Amazon Fraud Detector, see our Catching Fraud Faster by Building a Proof of Concept in Amazon Fraud Detector blog post.