<\/p>\n
Visa helped set the path for the payments industry by introducing a shift to EMV chip cards in 2011, which since has spurred a 75% drop in US card counterfeit fraud [1]<\/a>. However, success in reducing card fraud has moved the battleground into the digital space, as U.S. digital transaction (\u201ccard not present\u201d or \u201cCNP\u201d) losses grew 11% annually from 2015 to 2018, reaching $4.4 billion [2]<\/a>. Visa, a global payments technology company enabling fast and reliable electronic payments across a global network of merchants, financial institutions, and consumers, sits at the heart of this security battle and will remain a key participant in resolving widespread concerns around fraud, identity theft, and digital privacy. The continued development of machine learning capabilities may serve as a vital tool for Visa in better securing transaction networks.<\/p>\n Amid the mass of data points and contextual attributes (device identification, user behavior analysis, timestamps, payment histories) used to determine the likelihood of fraudulent activity, \u00a0legacy systems have relied on systemized rules to flag anomalies. To match increasingly complex attacks, these rules-based systems have grown into a layered confusion of tools and signals, no longer flexible enough to effectively manage the risk vectors. This ultimately has a detrimental impact on the customer experience of legitimate users caused by high numbers of false positives and negatively impacting revenue [3]<\/a>. Networks like Visa must build adaptive machine learning algorithms to deliver complex risk decisions quickly, securely, and seamlessly.<\/p>\n Visa has invested significantly in fraud prevention and analytics solutions in recent years to combat the escalating problem, including acquisitions of fraud and authentication platforms CyberSource (2010) and CardinalCommerce (2017) that together provide Visa, merchants, and issuers with a more complete end-to-end view of the transaction origin and authentication path [4]<\/a>. In the near term, the Company is continuing to integrate machine learning algorithms to improve accuracy, using a mix of ensemble models, supervised, and unsupervised methodologies. Given threat actors may also be increasingly utilizing machine learning as a tool to disguise their activities, Visa recognizes the value in its massive data sets as essential elements in properly training fraud detection algorithms [5]<\/a>. The Company will continue to integrate existing services such as Mobile Location Confirmation, a geolocation tool identifying if an account holder\u2019s mobile phone is near a purchase location, to enhance the analysis [6]<\/a>.<\/p>\n In the longer term, biometric signals may become a viable tool for Visa to further trim the potential risk, though introduction of biometrics presents a magnitude larger data complexity issue. Deep learning may provide a way to explore complex features within even larger data sets so that the model can learn better to predict frauds [7]<\/a>.<\/p>\n As Visa approaches integrating machine learning into its fraud prevention strategy, the Company might be best suited to enhance its solutions through acquisition or partnership. There are a number of smaller players in the fraud prevention space \u2013 representative vendors include Feedzai<\/a>, Signifyd<\/a>, and Sift Science<\/a> \u2013 that could benefit from Visa\u2019s scale of available data and that Visa could incorporate to speed up its technology development. Acquisitions by other competitive payment networks MasterCard (NuData, Brighterion) and American Express (Accertify) signal the market-wide impetus to seize the machine learning opportunity [8]<\/a>.<\/p>\n One area Visa should consider in development of these systems is their auditability. Any system relying so extensively on behavioral, tracking, and biometric data to create consumer identities may need to provide mechanisms for auditing its decision criteria and demonstrate the basis for decisions is lawful and ethical. Prioritizing a clear understanding of the premises for machine learning-enabled decisions will be important in establishing trust.<\/p>\n <\/p>\n Open Questions:<\/em><\/p>\n Blockchain technologies may become a more significant part of a secure transaction network. How might Visa\u2019s established network need to adjust for such an environment? What might these technologies do better than a data-centric machine learning approach to risk mitigation?<\/p>\n <\/p>\n (Word Count: 782)<\/em><\/p>\n <\/p>\n References:<\/p>\n [1]<\/a> Pegoraro, Rob. \u201cWhy those chips in your credit cards don\u2019t stop fraud online\u201d. Yahoo Finance<\/em>. November 5, 2018. https:\/\/finance.yahoo.com\/news\/chips-credit-cards-dont-stop-fraud-online-200041701.html<\/a>. Accessed November 2018.<\/p>\n [2]<\/a> Pegoraro, Yahoo Finance. <\/em><\/p>\n [3]<\/a> Care, Jonathan and Phillips, Tricia. \u201cMarket Guide for Online Fraud Detection.\u201d January 31, 2018. Gartner, Inc.<\/em>, accessed November 2018.<\/p>\n [4]<\/a> Visa Inc., September 30, 2017 Form 10-K (filed November 10, 2017), via Visa company investor relations, accessed November 2018.<\/p>\n [5]<\/a> \u201cMachine Learning in the Payments Industry.\u201d Visa Research. May 24, 2018. https:\/\/usa.visa.com\/dam\/VCOM\/global\/support-legal\/documents\/webinar-machine-learning.pdf<\/a>. Accessed November 2018.<\/p>\n [6]<\/a> Visa Inc., 2017 Form 10-K.<\/p>\n