Staying Ahead of Common Hotel Frauds with AI

With a rise in ecommerce transactions, combined with the growing threat of fraud in card-not-present (CNP) transactions, it’s more important than ever to get proactive and creative about fraud prevention.

If you haven’t yet incorporated AI into your fraud detection system, it may be time to consider it as a replacement for inefficient, error-prone processes that are probably resulting in too many chargebacks and costing you too much of your revenue.

For example, behind the scenes of Sertifi’s authorization solution are Kount’s fraud detection tools, which alert you to a potentially risky card before a guest’s arrival. As a merchant doing business online, it’s a great way to increase confidence that you’re accepting guests and cards you can trust, while maintaining a great customer experience and expanding your business with fewer chargebacks.

How can machine learning reduce credit card fraud?

Machine learning is an extension of AI that uses mathematical models of data to train a computer to make decisions without human input. In the case of detecting credit card fraud, computers analyze and learn from credit card data to make educated decisions about fraudulent activity.

Here’s how it works with the Sertifi / Kount integration:

1. An interaction happens. In this case, a person submits credit card information via a Sertifi credit card authorization form.
   
   
2. Data is collected, including card details, name, payer IP address, email, physical address billing, an AVS check, and hundreds of other data points.
   
   
3. Data is reviewed. The system analyzes the data using two types of machine learning models, supervised and unsupervised (more on that below).
   
   
4. Fraud risk is determined. The system looks for red flags that could signal fraud and determines the card’s risk level with an A-F score. The score is a summary of the card’s complete history.
   
   

Transactions receive D and F ratings only when there are serious indicators of risk. For example, the card will be declined if the system detects hundreds of bookings being placed from a single location/IP address, a warning sign that card testing fraud may be at play. For hotels, this means the fraudster books several rooms, and if they go through, they know they can use the card for additional purchases or sell it to another fraudster. If rooms are booked with no intent of keeping the stay, you miss revenue from the room and the card processing costs, plus real guests may be negatively impacted by very limited room options being available.

Using the score and reasoning, you can decide whether to move forward with a card or not, leaving the ultimate decision-making in your hands. Businesses have different levels of risk they’re willing to take on, so the fraud scoring is only intended to supplement your decision-making. For example, the system will flag a same-day check-in or if the card is being used in a different location than the guest. There may be perfectly safe and reasonable explanations for these scenarios, so even when a card is flagged as risky, we always recommend following up with the person who submitted it and seeing if it’s legitimate business you can generate revenue from.

What is supervised versus unsupervised machine learning?

• Supervised machine learning is the system’s memory, looking at historical patterns and trends to predict the outcome of the current interaction. For example, it recognizes an email address it’s seen in the past and remembers there’s a negative history with it. Over time, the system is getting more training and smarter about how to respond to particular information.
  
  
• Unsupervised machine learning is based on instinct. The system looks at the current attributes of the interaction and immediately connects the dots on where the person has been and the card history associated with them. For example, it analyzes the type of device and the email address being used.
  
  

Because unsupervised machine learning focuses on short-term linkages and patterns, it catches emerging fraud attacks and anomalies that supervised machine learning cannot yet learn about due to the recentness of unseen attack types. Unsupervised learning is also much faster and more accurate than human judgement alone.

How Unsupervised Learning Works

Source: Kount

 

Can machine learning systems help reduce chargebacks?

Yes, machine learning in fraud prevention systems can help reduce your chargeback rate – but keep in mind not all chargebacks are the same. Machine learning engines are trained to detect suspicious activity and help you reduce criminal-related chargebacks; however, not all chargebacks are due to legitimate fraud. Maybe your guest decided to initiate a chargeback instead of a refund for a legitimate transaction or genuinely forgot about a purchase they made from you. Machine learning would not prevent those types of chargebacks, though there are other steps you can take in these cases.

What are the benefits of using machine learning to prevent credit card fraud?

The biggest benefit is you’re replacing overly manual, error-prone processes with automation – without even losing any control. This makes it much easier to stay ahead of fraudsters and make data-driven, accurate decisions about who to trust. When you’re able to stop a fraudster early, you won’t see a chargeback eventually pop up, saving your revenue and your staff a significant amount of time.

In the end, machine learning is only as powerful as the data, support, and technology around it. That’s why Sertifi partners with Kount, a global leader that helps businesses stay ahead of evolving threats, seamlessly automate inefficient processes, and proactively protect revenue.

Plus, we can help you enforce other best practices at your property to reduce your risk of getting chargebacks from guests on legitimate transactions. Check out our detailed guide here.