arrow left facebook twitter linkedin medium menu play circle
April 14, 2021 - Chandreyee Chakravarty

How a Top Payment Network Beats Transaction Fraud

When you’re handling more than $8 trillion in payments each year, losing even just 1% of those transactions to fraud is a major threat. One top payment network was struggling to keep up with the ever-evolving tactics fraudsters were using. Because tactics evolve so quickly, the company experienced low detection rates and low accuracy, which left it vulnerable to high rates of transaction fraud and the potential to create friction for good customers.

Here’s how DataVisor helped the company flip the script and get a firmer grip on transaction fraud.

Time and Accuracy Created Big Fraud Detection Challenges

The payment provider was using traditional data models to find and prevent fraud. But because of the time (3-6 months) involved in tweaking and fine-tuning the models, the models were already decaying upon deployment. This left the company vulnerable to unknown fraud, as the models could only detect known types of fraud.

What’s more, the deployed models often experienced inefficiencies because the models could not incorporate key pieces of information. Specifically, many forms of digital data could not be included in the fraud models, rendering them less effective than they potentially could be.

Transactions happen quickly, so time is of the essence in determining fraudulent purchases. Real-time visibility plays an important role in this process, but the client’s current fraud approaches didn’t accommodate this level of responsiveness.

Enterprise Machine Learning Filled Important Gaps in Fraud Prevention Strategies

DataVisor’s Enterprise ML includes the use of unsupervised machine learning (UML), which has proven effective for this payment provider in multiple ways. For starters, UML isn’t limited to known fraud and data labels, allowing it to identify suspicious patterns that otherwise would be undetected. 

Also, UML doesn’t require time-consuming fraud modeling and refining, so real-time visibility became a possibility with DataVisor’s solutions. 

One of the main differences between supervised machine learning and unsupervised machine learning is that UML can compare more than one transaction at a time. It looks for relationships and patterns between multiple transactions. Suspicious indications, such as transactions occurring from a single IP address in a short timeframe, help to detect more fraudulent activities and with higher accuracy.

Part of DataVisor’s advantage lies in its ability to review data from more than 4.2 billion users and 800 billion events. With just a 50-100 ms latency, events are analyzed in real time across the enterprise so immediate action can be taken.

Combining SML and UML, the payment provider is able to significantly reduce model decay. Unknown fraud patterns not accounted for in supervised machine learning can still be detected, giving the client superior all-around protection against fraudsters.  

The Results: Improved Detection, Accuracy, and Relevance

As a direct result of implementing DataVisor’s Enterprise ML solution, the payment provider saw a 20% uplift in transaction fraud detection at a 94% detection accuracy. Building new fraud models is also 5 times faster with DataVisor, going from 4-6 months to just a few weeks, and those models do not immediately decay upon deployment.


See specific insights about DataVisor’s transformation for this payment provider; download the case study today.

about Chandreyee Chakravarty
Chandreyee is the Regional Head of Sales at Datavisor. She has held high-profile sales positions within the Identity Management and AI space over the last 20 years globally.
about Chandreyee Chakravarty
Chandreyee is the Regional Head of Sales at Datavisor. She has held high-profile sales positions within the Identity Management and AI space over the last 20 years globally.