Discover how some financial services firms are currently leveraging artificial intelligence and machine learning in fintech and regtech.
At the NSCP National Conference in November, I’ll be speaking on a panel with Jane Stabile and Andrew Siegel on the topic “Will AI Revolutionize the Financial Services Industry?” This is the second in a series of related blog posts on AI and its impact on the industry. Part 1 covered the current state of AI development in fintech. Here in Part 2, let’s look at how financial services firms are currently leveraging AI in their daily operations.
A couple years ago, Deloitte’s Insights series published an article announcing “the AI leg of the digital marathon.” The article noted that financial services firms on the cutting edge were already starting to implement “intelligent solutions such as advanced analytics, process automation, robo advisors, and self-learning programs.” But the article also recognized, at that time, we were a long way from widespread use of these technologies. Fast-forward two years, and we find that the deployment of AI in financial services firms is still in the early stages.
Where Are We Now?
EY recently published an infographic showing that most firms across all industries are still using “standard” tools to manage their businesses. For example, 95 percent of business owners still use spreadsheets or similar tools to manage their internal financial planning and analysis, while 57 percent use some form of business intelligence. “Emerging” tools include data visualization, reporting automation, cloud services, mobile applications, robotic process automation (RPA), and AI/machine learning. These last two tools, which will become the most disruptive in the industry, are used by fewer than 5 percent of firms.
So what’s the holdup? Part of the issue is data. Building AI/machine learning tools requires large data sets to train and refine the underlying machine models. Again, EY points out that “there has been a minimal shift in the time allocated to gathering data.”
Another reason for the slow spread of AI in the financial services industry is cost. Two years ago, 87 percent of Deloitte’s frontrunners in AI initiatives were spending over $500,000 per year on AI development, while a whopping 45 percent were spending over $5 million per year. The time, money, and resources needed to develop functional machine learning algorithms are substantial.
For example, IMP developed its CLEAR Compliance system using a type of machine learning commonly referred to as natural language processing, or NLP. IMP’s CLEAR Compliance system helps asset managers facilitate critical components of their investment trading compliance program by automatically reading prospectuses and identifying trading rules and restrictions. We asked Jon Gold, Managing Director at IMP, to tell us more about his experience with machine learning in fintech and regtech.
What Is NLP, Anyway?
Jon Gold: Arguably the most famous line from the 1968 film 2001: A Space Odyssey is this exchange between astronaut Dave and his renegade but polite HAL 9000 computer:
Dave: Open the pod bay doors, HAL.
HAL: I’m sorry, Dave. I’m afraid I can’t do that.
This was my first exposure to NLP—the intersection of AI and linguistics, allowing computers to understand and decipher language. We are perhaps unwittingly familiar with NLP, as it powers predictive text, spam filters, instant translation, and, of course, “Hey Siri!” The reading comprehension version of NLP is commonly known as natural language understanding (NLU), and this is where our industry can stand to achieve the greatest gains in efficiency.
We haven't yet reached a world where a client can say, “Hey Siri, tell my manager not to invest more than 10 percent of net assets in any single industry—including cash.” This edict, like much of the client onboarding experience, is driven by paper. At IMP, we’ve been using our proprietary NLU application on the volumes of documents that underpin both the onboarding and the risk and compliance functions to help our clients achieve greater efficiency, higher rates of accuracy, and, perhaps most importantly, enhanced auditability.
How Can NLP and NLU Make Operations More Efficient?
Jon Gold: Let’s begin with the “four eyes” standard. This longstanding internal control mechanism mandates that activities or decisions which create potential risk for the organization are reviewed by a second individual who is capable and independent. What could be more independent than an algorithm? Over time and training, NLU will extract less verbiage and more key words and phrases, thus better deciphering the intent of the language.
As we know, a sizable proportion of the language of investment management is boilerplate. The “good stuff” to which you’ll be fiduciarily obliged to adhere is often buried in reams of text. Unlike humans, algorithms don’t “skim,” so the risk of having missed something important is reduced. Can your third and fourth eyes claim that?
Each time a document passes through the model, the human trainer evaluates its output. That output is language which the algorithm thinks is important. A simple “yea” (yes, this is a good find) or “nay” (no, not something we need to capture) is how the algorithm learns. The greater the volume of documents consumed by the model, the more discerning it becomes as it gets more adept at recognizing the language of the industry and of the firm. The number of “nays” is continually reduced.
This capacity will liberate more experienced staff, whose expertise will be manifested in the algorithm. In an environment where staff are compelled to do more with less, time-critical tasks such as overriding a stopped transaction can be attended to more promptly. There’s no way to quantify the savings from reducing the latency of an in-flight order, but the risk of missing the market is certainly reduced.
NLU is useful not only for parsing out important passages from a document on its initial read. The same machine learning can be applied to a document comparison when agreements are renewed, regulations are updated, or other language changes. This goes beyond the standard text matching process with which we are familiar in word-processing applications. Because the validated output from the NLU algorithm links directly into specific requirements (say, a notification procedure or a compliance test), it can instantly invoke tasks that need to be completed or restrictions that need to be monitored differently.
What’s the Catch?
Jon Gold: The advice columnist Eppie Lederer (better known as Ann Landers) is credited with the quote “If you think education is expensive, try ignorance.” Training any algorithm is a process without shortcuts. If the training data is inadequate, or if the feedback is poor, achieving positive results will be slower. Couple that with the instant gratification expectations of the investment industry and you're likely to experience disappointment before success.
But you don’t start a new hire and expect them to begin firing on all cylinders from day one. AI models are no different, except that their capacity for throughput is considerably greater than that of any human being. We mere mortals are still a key part of the formula, because we can't train an algorithm without sharing our unique analytical reasoning skills with a green computer. There’s no doubt that perseverance does pay off, yielding an infrastructure that can accommodate greater volumes of business.
Where Are We Going?
While consulting firms like IMP are using machine learning to develop specific solutions for financial services firms, other fintech companies are taking an aggregator approach. For example, digital banking unicorn Revolut offers bank accounts, debit cards, fee-free currency exchange, stock trading, cryptocurrency exchange, and peer-to-peer payments. Revolut’s products leverage APIs built by other banking and financial services companies. As a fintech aggregator, Revolut focuses on creating a smooth, user-friendly experience through its applications. Revolut centers its technology on improving personalization and providing financial management tools that aren’t offered by other fintech companies in a single application. In other words, Revolut focuses more on integration than innovation, with its primary value proposition being the user experience.
Stay tuned for more blog posts in this series on AI in fintech and regtech. In Part 3, I’ll provide some practical advice on how your firm can start an AI project. And in Part 4, I’ll wrap up the series by discussing ethical implications of AI development and implementation.
In the meantime, Joot is assembling an advisory board of tech-savvy investment advisers interested in codeveloping AI-driven regtech tools. If you'd like to participate, please indicate your interest here and we'll be in touch.