Sign up for The Paypers newsletter Follow The Paypers on LinkedIn Follow The Paypers on Twitter Follow The Paypers on Facebook
The Paypers, paypers, Insight in payments, News, Reports, Events
 advertisement
Case study

US credit card issuer uses DataVisor machine learning to reduce losses from application fraud

Wednesday 20 February 2019 | 10:11 AM CET

With the rapidly-increasing application fraud, DataVisor’s machine learning solution enables a top US credit card issuer fight this type of fraud efficiently

Quick Snapshot

Challenge:

  • Third-party and synthetic fraud caused huge financial loss

  • Legitimate applications were disrupted resulting in poor customer experience

  • High manual review rate and operational cost for fraud teams

Solution:

  • Proactively captured entire fraud rings by utilizing machine learning models to identify patterns across new applications rather than evaluate them in isolation

  • Improved good customer experience with low false positives and high detection accuracy and boosted operational efficiencies by enabling bulk decisions for the entire fraud ring

Results:

  • Captured 25% additional fraud and saved over USD 15 million fraud loss and operational cost

  • 94% detection accuracy with only 0.17% false positive

DataVisor recently partnered with a top US credit card issuer who offers a variety of credit products to help consumers finance purchase of goods and services. The issuer processes over USD 20 million applications every year and it was looking for an effective and scalable solution to fight the rapidly-increasing application fraud, especially coordinated fraud from the online channel.

Capture coordinated third-party and synthetic fraud

Application fraud was on the rise and had become a significant challenge to the customer as a large number of fraudsters operated using stolen or synthetic identities, and recruited people to open accounts. Their attacks were highly professional and coordinated. The customer did not have a systematic way of finding linkages among different fraud incidents and detecting fraud patterns in real time. Their existing tools captured only a portion of that fraud ring.

Proactive and agile fraud detection

Prior to using the DataVisor solution, the customer was using internal rule systems and multiple supervised machine learning (SML) models for fraud detection. However, the rules were reactive and had to be frequently updated to capture new patterns. Also, the SML model relied on labelled data for training and the labels usually wouldn’t arrive until months later. The accuracy of labels was also questionable since fraud defaulters were often mixed with credit defaulters. As a result, it usually took the customer at least 6-12 month to refresh a supervised machine learning model and it was still not very accurate. The card issuer was looking for a proactive and reliable solution to help them get ahead of the fraudsters.

Balance risk with CX and efficiency

At the same time, the customer was looking to manage risk while keeping customer experience intact. For each genuine customer reported as suspicious, it took the customer from several hours to two weeks to open the case, perform manual reviews, and take action to verify the applications which created friction for them. The customer did not want their legitimate customer applications to be disrupted.

Additionally, their operational cost was on the rise because of the large volume of alerts and a high false positive rate. Their operational team was overwhelmed by alerts and queues that required intensive manual reviews. Moreover, there was a critical need of a solution that would significantly reduce time to review alerts and manage their work efficiently.

Immediate ROI and high accuracy with machine learning solution

The card issuer considers DataVisor Enterprise the best complementary solution for significantly improving operational efficiency and for enabling the business to capture the entire fraud ring. Without the need for fraud labels, historical data or extensive training period, they integrated DataVisor Enterprise in a few weeks and started seeing results right away; its solutions immediately detected fraud that bypassed the issuer’s other solutions and provided real-time scores that were highly reliable.

The customer found DataVisor’s unsupervised machine learning solution very powerful in finding hidden connections amongst incidents, identifying new attack patterns including very sophisticated fraud rings. DataVisor Enterprise not only analysed profile information, but also took a holistic approach to view cross-account linkages, behavioural data and digital footprints such as device IDs and datacenter IP prefixes and browsers.


By performing trends and pattern analysis, DataVisor’s machine learning model captured 25% more fraud in addition to what the internal team detected. The solution also caught fraudsters early on at the application stage -- usually two days to a week earlier than other solutions -- saving the card issuer over USD 15 million every year.

Enhanced operational efficiency and confident decision-making

The operational team vastly reduced the alert volume and improved their efficiency by working with DataVisor to define the threshold for different actions. For highly risky cases, the customer had the option to automatically route them to friction heavy queue with DataVisor. For less suspicious cases that required human investigation, DataVisor discovered clusters of fraudulent accounts and group results; therefore, analysts needed to review only ten to twenty sample cases and then confidently make bulk decisions that apply for hundreds and thousands of cases in the same fraud ring.

With 94% detection accuracy and only 0.17% false positive, the card issuer developed high confidence in DataVisor’s results enabling them to acquire their customers aggressively without worrying about fraud risks. In the long term, the customer expects to extend DataVisor’s machine learning solution to their various products - consumer and small business for both online and offline applications.

About Fang Yu

Fang Yu is the Cofounder/CTO of DataVisor, where her work focuses on big data for security. Fang has developed algorithms for identifying malicious traffic including fake and hijacked accounts, and fraudulent financial transactions. Fang received her PhD from UC Berkeley and holds over 20 patents.

About DataVisor

DataVisor is the next gen anti-fraud platform based on cutting edge AI. Using proprietary unsupervised machine learning algorithms, DataVisor helps restore trust in digital commerce. Combining an intelligence network of more than 4B user accounts globally, the DataVisor solution is deployed across a variety of industries, including financial services.

 advertisement
 advertisement
 advertisement
 advertisement