Artificial Intelligence in Payments
We have started development on this year’s Innovation in Payments workshop, and I wanted to share our current thinking on a new topic we’re including for the first time –– the use of artificial intelligence technologies in the payments industry.
AI technologies have been around for a while in the payments industry –– think about the neural networks used for dynamic risk scoring –– but it has never been hotter as a topic than it is right now. This is largely due to the huge amounts of venture financing that is flowing into AI, the emphasis that Google and others are putting on AI as the next market battleground, and the tangible results that have been shown to date.
The investments are significant. According to CB Insights, over 200 companies raised $1.5 billion of equity funding in the first half of 2016. Most of that is going into self-driving cars, enhanced medical diagnosis, and intelligent assistants. But some of those investment dollars are also flowing into machine learning startups dedicated to applying AI techniques to the world of finance.
But before we get too far into the use of AI in the payments industry, let’s circle back and explore what people are talking about when they use the term “artificial intelligence”. There are many definitions that people use, but the one that seems to resonate best with how AI technologies are used in the payments industry speaks to the ability of algorithms to learn and adjust based upon changes across the payments ecosystem. This is possible, in part, because of the feedback loop inherent in payments. Bad transactions at the issuer level are reported back to the network, as specified by the operating rules, where they can feed the risk scoring algorithms. Bad transactions at the merchant level, for example, get reported back to the merchant in the form of chargebacks.
AI specialist tend to break the discipline into a number of well-established domains such as natural language processing, vision recognition, robotics, and decision making. I haven’t seen the payment robots, yet, but there are plentiful examples of payment and commerce-related applications in the other domains.
The domain of natural language processing is the home of automatic speech recognition (e.g., Apple Siri), text-to-speech conversion (e.g., Audible playback of books), and automated language translation (e.g., Google Translate) among other areas of specialization. In the world of commerce, Amazon’s Alexa is now able to recognize audible commands to buy products from Amazon and its partners. More payment-oriented still, Apple has extended Siri to support a vocabulary for initiating P2P payments through voice commands. Square Cash, Venmo, and others are using this capability today.
Vision recognition as a domain is the home of object recognition, facial recognition, event detection, and motion tracking. Of these specialization, facial recognition seems to have the best fit in the payments industry. Jack Ma, CEO of Alibaba, recently demonstrated “smile to pay” which is essentially the use of facial recognition as a second authentication factor to unlock payment credentials and initiate a payment. Mastercard is working on a similar concept that it calls “Mastercard Identity Check”. Google is also developing a POS payment capability that uses facial recognition. In the popular press, the use of facial recognition in payment authentication is often categorized to as a “selfie pay” innovation. I love that.
Decision making is by far the most mature of the AI domains, with every day areas of specialization like product recommendations (e.g., Netflix Recommendations), scheduling optimization (e.g., airlines), and route planning (e.g., Google Maps). In the world of commerce and payments, there are areas of specialization in card fraud detection, card portfolio optimization, offer personalization, and money laundering detection.
The use of AI technologies in automated card fraud detection is important for a number of reasons, not the least of which is the sheer magnitude of the amounts of money moving through the card system. And while global card fraud absorbed by issuers, acquirers, and merchants reached $21.84 billion in 2015, according to The Nilson Report, it only represented 7 basis points of losses in total.
Card issuers have long used the FICO Falcon Fraud Platform for automated risk scoring of every card authorization request against a pool of billions of card transactions. On the merchant side, machine learning is increasingly being used alongside address verification, device IDs, and experiential databases to mitigate the impact of online fraud. Companies such as Sift Science and Feedzai have brought a pure machine learning approach to the established field of online risk management. The traditional providers such as CyberSource and ACI ReD Shield have also added machine learning technology to their suite of risk management tools. Stripe recently introduced a machine learning technology it calls “Stripe Radar” to augment its traditional approach to risk management.
At a macro-level, AI-based innovation is flourishing in the payments industry. Beyond the specific examples mentioned, there is a growing segment of the industry focused on offer optimization. There is also early thinking being done on the use of machine learning to dynamically determine the optimal routing path for transaction authorization. Interesting stuff.
To learn more about the use of Artificial Intelligence in the Payments Industry, I invite you to attend Glenbrook’s upcoming Innovation in Payments workshop being held December 8th in Palo Alto. This is a special Glenbrook Insight Workshop being held after our final Payments Boot Camp of the year. If you are interested in attending both, there is special discounted pricing available when you bundle both workshops together. I hope to see you there.