Retooling for Fraud Detection and Prevention
In our previous post, we discussed how lack of clarity and focus prevent businesses from realizing ROI for AI. To combat this, we recommend businesses identify a clear use case within an operational task. Today we’re going to look at a few opportunities within fraud detection and prevention. Fraud is a great place to start because incremental improvements directly improve your bottom line by way of reduced defaults and fraud losses. In addition, when AI is implemented properly, you can reduce a lot of the operational costs associated with fraud mitigation. LexisNexis’s fraud study found that in financial services, every $1 in fraud losses equates to almost $4 in total cost.
However, fraudsters commit different types of fraud and take different entry points to infiltrate your business, each requiring its own strategy for fraud mitigation. Therefore, we recommend focusing on one at a time. According to the Consumer Finance & Lending Sentiment Survey we conducted last fall, application fraud is the most common fraud type that FIs are experiencing. So let’s explore how AI can be applied to improving application fraud detection and prevention.
- Identifying patterns from your existing customer data. Assuming that you’re capturing known cases of fraud in your database, finding linkages between application data and fraudulent customer data can help you detect fraud. For example, if the phone number of a new applicant matches the phone number of a known fraudster, this new applicant may be fraudulent. If you’re not capturing known cases of fraud in your database, that’s a great first step. In the meantime, finding linkages between application data and all existing customer data can be as useful. For example, if the phone number of a new applicant matches the phone number of an existing customer and the address of that new applicant matches the address of a different existing customer, this new applicant may be fraudulent. And in fact, you may have identified a case of synthetic identity fraud, where a fraudster pieces together real information from different people to form a fake identity.
- Identifying patterns from aggregate behavior. Assuming that you’re capturing the metadata around the application, finding patterns of behavior that deviate from the norm can help you detect fraud. For example, if you tend to see very few applications come in during the late evening but on one day an influx of applications come in at around 2AM, these applications may be fraudulent and you may have been victim to a cyber attack. If you’re not capturing the metadata around the application, that’s where you should start.
- Identifying patterns from consortium data. Why limit yourself to your own data? Leveraging consortium data enables you to learn from your peers. For example, a fraudster may apply using a device that your business has never seen before. Assuming that there are no other red flags, you will most likely let this fraudster through. Little did you know that this device was associated with fraud at several other companies. If you had access to consortium device data, you would not have let this fraudster in.
Applying AI in a manner that enables you to identify these patterns and take action quickly will lead to improved fraud detection, reduced operational costs, and minimal customer friction. Now that we’ve discussed possible use cases, here comes the age-old question: To buy or to build? We’ll devote a whole post to addressing that question later on in the series. As a sneak peek, we like what McKinsey has to say about ecosystems.
Want to get up-and-running faster?
Contact us today to learn how you can take advantage of our application fraud detection solution accelerator.