The Role of AI in Fintech Disruption

(This is Part 2 in the series The Next Wave of Fintech Disruption. You can find Part 1 here: )

AI is going to change everything. Throw out the old paradigm, in with the new. Every coffee shop in Silicon Valley is abuzz with talk and excitement about the potential of AI.

2017? No, 1987. Thirty years ago saw the first hype cycle about how Artificial Intelligence, with all the excitement, the drive to learn more and incorporate that into personal career growth, the confidence that it would be The Next Big Thing and would turn both the current computing paradigm and the established world business order upside down.

Deja Vue. Or maybe back then it was Presque Vue, a past vision of now…because AI is imdeed all the rage again. Pity the poor startup today that does not include AI in its pitch deck, it’ll have a hard time getting an investor to even take a meeting. But the hype and excitement today is based on bigger and more profound changes that are pervading every computing process. If software is eating the world, then AI is eating software, as some have said.

The technology basis for AI today is vastly different than in the first go round. Back then it was primarily technology that could allow the programming of an “Expert System” that would use a rules hierarchy to implement automated categorization and decisions. A spate of languages were developed to allow the programming of these systems, and when Borland (a software publisher that had a big role in democratizing programming by offering programming tools at $59 instead of $750…the most famous being Turbo Pascal) released Turbo Prolog some thought it was going to be an historical inflection point.  So much for that. (It is worth noting though that some of today’s most important AI technology is directly rooted in other efforts back then, so it wasn’t like all of yesteryear’s technology got tossed.  For example, academics including James McClelland and David Rumelhart at UCSD began publishing a series of papers that laid out the foundations for Neural Networks, which are the foundations for today’s Deep Learning, Random Forest, and related areas of AI.

AI today comprises a huge variety of sub fields, and sub-sub fields. There are the neural network based disciplines, there are fields based on Bayesian and non-Bayesian statistics, there are techniques that have evolved from the “Search” capabilities of ten or fifteen years ago, and there are rapidly evolving approaches to layering these sub fields together. IBM has done a great job of branding Watson as a single, omniscient intelligence, but under the covers it is really a collection of many—over 100—technologies to handle the various steps in arriving at a good answer.

Not surprisingly, there is no standard argot or terminology to cover all this, yet….what one person may call Data Science, another might call Machine Learning, or Decision Theory. Layer in terms that classify techniques according to e.g. whether they are  Supervised Learning rather than  Unsupervised Learning and you have a situation ripe for confusion--or obfuscation. (A good overview of all this is contained in this recently released JP Morgan Big Data and AI white paper.)

If the variety of AI technologies is challengingly diverse, the applications of AI into Fintech Disruption are just as eclectic, and are starting to show real promise for changing the way that financial institutions (including regulators) do business, and the way that Fintechs can drive change in Finance.  But the most important way to understand the advent of AI is that it will be ubiquitous in every dimension of financial services computing. And so nearly every Fintech startup…whether in payments, crowdfunding, robo, regtech, Insuretech, mobile banking, customer experience, customer acquisition,  STP improvement, etc… is looking to employ “AI” in some capacity. This leads to two interesting points. First, Don’t Be Beguiled by the AI…if every Fintech is going to have it, then FIntechs won’t be able to differentiate or succeed just by saying they have AI…they have to prove core value, and prove themselves by delivered ROI and benefits. (The AI is just today’s hygiene, in a sense.) Second, Don’t Be Beguiled by the AI…most FIntechs are saying they have AI as a come along…hiking up their skirts, so to speak…but if you drill into the size and composition of their technical teams, and what they have available for AI operations and model development, you’ll get insights into their challenges at making AI actually work to deliver results.

This is not a criticism, because it’s early days in AI…the Beginning of the Beginning, as some have said. And so while a startup may know that its manifest destiny is to employ AI, and may be working hard on developing a particular capability, there may be a lot more manual or human involvement today than you would think, from the hype. But today’s operations will lead to the understanding that will enable automation, and so in a sense is a necessary phase. Meanwhile, the “democratization” of AI is tearing along….less expensive and more widely available teaching and training on AI, more integration into CS and professional (“boot camps”) curricula, public sharing of accomplishments and techniques, more and more open source code and better and better programming and computing infrastructure. And so Fintechs’s trust and faith that they have an AI future is well grounded, as it becomes easier/quicker/cheaper to incorporate AI.

Existing institutions such as banks and hedge funds are not sitting around to leave AI-driven business innovation up to upstart Fintechs. Any enterprise with a big enough IT team is trying hard to cultivate the ability to create and utilize (whether homegrown, or from a partner) AI. And in this regard, even banks that have been around for decades or centuries face a same fundamental challenge that young Fintechs do: Data Makes the World Go Round. Basically all AI techniques are completely and fundamentally dependent on data, and often in multiple ways. It takes data to “train” a model (build up a computational capability to assess a context or decision) and it takes data to continually re-calibrate the model and keep it viable as the real world changes.

So you need data, and usually the more, the better, especially for neural network derived approaches. Google’s team that did the groundbreaking work on recognizing cat images on the Internet wasn’t finding success until it tried scaling the training data set up to something like a million images. And though other techniques may require a lot less data –modeling techniques that are basically multivariate regression analysis rather than Deep Learning may need only hundreds or thousands of data points – more is generally better. You may have run into the effects of this, yourself, in interactions with credit card company’s fraud detection systems. If you have traveled consistently over ten to fifteen years using your personal card, that company’s fraud system (which after all is an AI) is probably much better at recognizing that your restaurant expense in Tokyo was legitimate, than the system behind your corporate card, which depending upon how long you’ve been with your employer may have only 2-3 years of data upon which to base its judgements. (Though in truth these fraud models rely upon many other sources of data as well, not just your events.)

This dependency on data presents an impediment to AI attainment by Fintechs, because often they are challenged to find enough data to train, evolve, and improve their models. This is particularly true in more esoteric finance use cases, where there just may not be enough good training data available for purchase, or if there is it is prohibitively expensive to an early stage startup. And so there is a natural partnering motivation, between existing institutions, and Fintechs. (Watch this space for an upcoming post in this series, exploring these Institution/Fintech partnering dynamics in more detail.)

Credit card fraud prevention systems give us another important way to think about the state of AI in Fintech disruption, which is that machines and humans are (at least for now) co-dependent, and having an effective human-machine interface becomes a critical component of getting to success and ROI with AI. Nervous credit card companies have dialed up their AI to block your card at the drop of a hat but they’d spend a lot more in handling angry customers if they hadn’t developed smooth flows for SMS alerts and responses regarding the legitimacy of transactions, leading to reactivation of a card. And so in many cases the disruption that Fintechs bring to the market is essentially innovation in this human-AI interface, or in new business operations capabilities that allow banks to create new forms of human-AI interaction.

Disclaimer: The foregoing are my personal opinions solely, and are not the opinions or policy or advice of my employer, either express or implied. This post is not made in any official capacity, but solely as a personal activity. Citation of a particular company is not an endorsement or recommendation or criticism in any way.