Imagine if you could tackle fraud by leveraging the business intelligence of one of the largest transaction data sets in the world. Trained on Mastercard’s anonymized and aggregated global transaction data, Brighterion’s new market-ready artificial intelligence (AI) and machine learning (ML) models are revolutionizing fraud prevention and credit management.
Financial institutions (FIs) face fraud and credit risks every day. Nine out of 10 acquiring banks reported transaction fraud increased during COVID-19, according to PYMNTS.com. Meanwhile, U.S. lenders are doing business amidst rising interest rates – with household debt at an all-time high of $15.84 trillion. Now, these problems can be managed quickly and accurately with out-of-the-box AI solutions. These models are production ready to begin global deployment in as little as a few days.
With Mastercard’s global network of 210 countries and territories, the breadth of transaction data is vast. Using transaction data for financial data analytics while respecting customer privacy is a core value for Mastercard. All Mastercard transaction data has been aggregated and anonymized when used to build Brighterion’s AI and ML models. FIs will benefit from models trained on this data to learn patterns and anomalies. These models can instantly score transactions to deliver intelligence in real time.
Turning AI expertise and transaction data* experience into the future of AI for fraud and risk
Brighterion AI has proven its accuracy in preventing transaction fraud on the Mastercard network. With over 20 years’ experience in AI innovation, Brighterion strategized a way to develop faster-to-implement models. These models leverage Mastercard’s 50 years of payments experience and use current anonymized and aggregated data to enrich the models.
Pre-trained AI models don’t have to refer to network data during transactions or to score events. This enables the cloud-native platform to deliver a remarkably low latency of 100-150ms and, when deployed on-premises, a speed of less than 10ms.
What problems can global transaction data* help solve?
Brighterion offers market-ready solutions for credit risk management and acquirer fraud.
AI for credit risk
Leveraging anonymized and aggregated data from the Mastercard network, Brighterion helps lenders monitor for potential credit risk. Brighterion’s Transaction Credit Risk solution provides lenders with a real-time score for borrowers as it updates with each transaction fed into the model.
The solution’s early warning score predicts current accounts in good standing that are likely to become delinquent by recognizing anomalous patterns.
Banks can use the transaction credit risk score in conjunction with their existing models to increase prediction power. The model is designed to mitigate charge-off losses and reduces collection costs by ranking accounts likely to flow into more severe delinquency. Lenders can approve low-risk transactions that would have been declined due to insufficient funds for good customers.
One Brighterion customer achieved a 12 percent improvement in delinquency detection rates when combining the transaction credit risk score with existing risk models.
Market-ready acquirer solutions
These more timely, broadly trained models are also beneficial to acquirers and processors. Both transaction fraud and merchant risk increase with economic pressures, exposing acquirers to potential losses.
Global transaction data* brings knowledge from the broader market to deliver fewer false positives and higher fraud detection rates. Using Brighterion’s market-ready AI, one acquirer saw an increase in fraud detection of two to three times and an increase in approval rates of 7.4 percent.
Acquirers can monitor merchant transaction patterns to identify defined business risks while reducing friction for trusted users. The models provide a 360-degree view of the acquirer’s merchant ecosystem, covering risks such as fraud and collusion.
Acquirers can benefit from reducing their liability and save money by detecting fraud earlier in the payments flow. When implemented in the pre-authorization stage rather than at the transaction stage, the solution identifies the fraud before it hits issuer or network defences. Because the acquirer solutions are trained on the most sophisticated fraud attempts, protection and accuracy are high. The result is fewer false declines and fewer charges from issuers and network providers.
- Increase accuracy of fraud prediction rates
- Decrease fraud rates to protect accuracy ratings
- Prevent the most difficult to catch fraud
- Save money that would be lost to transaction costs, merchant fees and chargebacks
Customers are reporting a 15 to 30 percent lift in detection rates and a two-to-one false positive rate.
How does training AI on Mastercard transaction data* differ from using an FI’s own historic data?
Building traditional/custom AI models
Traditional, custom AI models require longer turnaround times. Some FIs struggle to extract the right financial data for model building and training or may not have the historic data that developers need.
FIs are burdened with extracting huge datasets to meet this need, ensuring correct labeling and effective data transfer. The required datasets may be in the hundreds, requiring a substantial time investment by the FI. These are used to train the new model to identify anomalies relevant to the business’ specific challenges.
The custom model is then built over the course of six to eight weeks, including testing, before it is ready for deployment.
Initializing self-learning, market-ready AI models
Market-ready models are built with a variety of advanced AI and ML technologies. These turnkey solutions move beyond the intelligence found in an FI’s own historical data; they are trained with the business intelligence derived from more than 150 billion transactions a year processed by Mastercard. By using this more robust data set, the model has expanded knowledge.
The key advantage of market-ready AI is that it saves FIs time and resources as the model is already built, trained and reflects high accuracy rates. The model is ready to deploy with a small sampling of the FI’s own transaction data. During this period, the customizable API interface is fine-tuned to the customer’s needs. Scores are then returned to the FI and the model is ready for full deployment.
For Transaction Credit Risk customers who are already Mastercard customers, deployment is even simpler as Brighterion can readily access the aggregated and anonymized network data.
That said, market models for both acquirers and lenders are brand agnostic. It is the scope of business intelligence that FIs benefit from, not the network brand involved.
As with Brighterion’s custom AI models, market-ready models are self-learning. They continue to learn and evolve with each new transaction or output with oversight by Brighterion computer scientists.
A place for custom AI
Brighterion has the process of building custom models down to a science – preparing customers for deployment in a couple of months with its AI Express implementation process.
Custom built AI models don’t have to be complicated and are still the right fit for specific or unique business challenges where innovation and experimentation is needed.
Do market ready models spell the end of custom AI? No, but for certain use cases, market-ready AI can save FIs time and money.
Market-ready AI models trained on Mastercard’s anonymized and aggregated global transaction data are another Brighterion innovation that continues to improve results and save customers time and money. These new models offer the advantages of globally accumulated business intelligence to deliver low latency and high accuracy. Currently available for transaction fraud, merchant risk monitoring and credit risk management, Brighterion’s market-ready models are brand agnostic and ready to operate on most FIs’ systems.
Learn how Brighterion’s market-ready AI is production ready out of the box to deliver superior financial analytics and immediate return on your investment.
*All Mastercard transaction data has been aggregated and anonymized when used to build Brighterion’s AI and ML models.