FX International Payments
By Mike Faden
Machine learning is being applied alongside existing fraud detection systems, which typically use manually created rules and other techniques, such as flagging unusually large withdrawals or payments initiated outside a cardholder’s home country.3 Machine learning differs from these traditional techniques in that it analyzes large amounts of historical transaction data to build a model that can identify patterns associated with fraudulent transactions. The system then uses this model to scan incoming payments in real time and flag potentially fraudulent ones.4
Machine learning is becoming increasingly important for several reasons, experts say. One is the increasing volume of e-commerce and other remote “card not present” transactions, in which speedy approval is often required yet the purchaser’s physical card is not present for additional verification.5 A 2016 report from industry association Financial Fraud Action UK found that online fraud against U.K. sellers increased 13 percent in 2015, and that remote fraud accounted for 70 percent of total losses.6 Remote channels also were primarily responsible for driving an increased level of fraud at U.S. merchants, according to a 2016 LexisNexis fraud study.7
Some observers say that another reason for the growing importance of machine learning in fraud detection is the worldwide shift to immediate payment systems, which require correspondingly faster identification of potentially problematic transactions.8Yet another is fraudsters’ ability to continually change their tactics to evade anti-fraud controls; fraud-detection systems must therefore continuously adapt to keep up.9
Machine learning is suited to addressing fraud in payments solutions in part because of the feedback loop inherent in payments, observes Russ Jones, a partner at payments industry strategy consulting and research firm Glenbrook Partners, in a blog post.10 When fraud attempts succeed, the bad transactions are continuously reported back to the payment network and can be fed into risk-scoring algorithms along with all other data associated with each transaction.
This means machine learning systems can analyze vast amounts of historical data to identify patterns associated with fraud. Experts say that machine learning technology is capable of taking into account many more data points than would be possible with manual methods alone, including detailed patterns of behavior associated with specific accounts.11 This may help the technology make a more accurate determination of whether a payment is likely to be fraudulent.Machine learning is coming to the fore now, technology providers say, in part because faster, low-cost computers and data storage make it possible for machine learning systems to process high volumes of transactions in real time, making decisions based on complex criteria in a fraction of a second.12
By helping businesses and payment services more accurately spot potential fraud, machine learning may provide several benefits. These go beyond reducing the losses that are directly due to fraudulent transactions; especially when selling online, businesses may be likely to incur additional charges, including fees, for payments that were accepted but subsequently determined to be fraudulent. 13 The LexisNexis study found that every dollar of fraud losses in 2016 cost merchants a total of $2.40 in chargebacks, fees and merchandise replacement.
Some machine learning systems can be used directly by businesses to scan transactions before they are submitted to payment networks, potentially reducing the incidence of such chargebacks.14
A second potential benefit is a reduction in manual review of transactions. This may be significant: the LexisNexis study found that large companies selling through online and mobile channels spend up to 25 percent of their fraud mitigation budgets on manual reviews.15
If machine learning decreases the number of times that a payment service incorrectly identifies valid transactions as potentially fraudulent (sometimes known as “false positives,” “false declines,” or “false alarms”) it may also reduce lost sales, customer frustration, and the associated potential reputational damage to the seller, experts say.16 According to IBM, one leading bank achieved a 40 percent reduction in false alarms on e-payments by applying machine learning technology. 17
Businesses and payment services are increasingly using machine learning technology to quickly distinguish between fraudulent and valid transactions, particularly as e-commerce volumes continue to increase. When used in conjunction with more traditional risk-management methods and manual reviews, machine learning may help to more accurately spot fraudulent payments before they are approved, potentially cutting costs by decreasing chargebacks, reducing manual reviews, and improving sales.
Mike Faden has covered business and technology issues for more than 30 years as a writer, consultant and analyst for media brands, market-research firms, startups and established corporations. Mike also is a principal at Content Marketing Partners.
1. “Why machine learning detects payment fraud more accurately,”The Paypers; http://www.thepaypers.com/expert-opinion/why-machine-learning-detects-payment-fraud-more-accurately/766108 .
2. “Using machine learning and stream computing to detect financial fraud,” IBM; https://www.research.ibm.com/foiling-financial-fraud.shtml .
5. Real-Time Fraud Detection – In Your Next Pizza Order, RTInsights.com; https://www.rtinsights.com/chargeback-fraud-prevention-ecommerce-delivery-fortent/
6. “Why machine learning may help stop payment fraud,” CIO Australia; http://www.cio.com.au/article/593627/why-machine-learning-may-help-stop-payment-fraud/
7. "The role of machine learning in real-time fraud detection," NCR Financial Blog; https://www.ncr.com/company/blogs/financial/role-machine-learning-real-time-fraud-detection/
8. "Artificial Intelligence in Payments," Payments Views; http://paymentsviews.com/2016/11/03/artificial-intelligence-in-payments/
10. The role of machine learning in real-time fraud detection,” NCR Financial Blog; https://www.ncr.com/company/blogs/financial/role-machine-learning-real-time-fraud-detection
12. “Machine Learning in Fraud Management,” Payments Views; http://paymentsviews.com/2016/07/07/machine-learning-in-fraud-management/
13. “A primer on machine learning for fraud detection,” Stripe; https://stripe.com/radar/guide#credit-card-fraud
14. “Overview,” Sift Science; https://siftscience.com/products/payment-fraud#overview
15. 2016 LexisNexis® The True Cost of Fraud Study , LexisNexis; http://solutions.lexisnexis.com/17057
16. “How machine learning can help FIs prevent fraud,” banking.com; http://banking.com/analysis/machine-learning-can-help-fis-prevent-fraud/
17. “Using machine learning and stream computing to detect financial fraud,” IBM; https://www.research.ibm.com/foiling-financial-fraud.shtml