It’s the age of algorithms, but not all algorithms are the same — and not all of them constitute true AI. A new PYMNTS report finds that FIs have adopted various forms of machine learning, but that AI’s deployment remains low. What will it take to get more AI involved in fraud prevention and other tasks? What are the long-term costs of settling for lower capability algorithmic technology? Get your algorithm on and have a read.
Algorithms are taking on more of the data and security work for financial institutions (FIs), with technologies such as data mining and business rules management systems (BRMS) finding popularity among banks and credit unions. However, fewer institutions have made the move to true artificial intelligence (AI), with funding and even misunderstanding about the technology serving as challenges to the wider acceptance of AI.
A new PYMNTS report entitled “The AI Gap: Perception Versus Reality In Payments And Banking Services,” done in collaboration with Brighterion, dug into the reality and hype of AI. “Artificial intelligence” has become one of those buzz phrases that mean much more than the reality, depending on who is doing the talking or providing the hype.
As the report noted repeatedly (and as AI experts keep reminding the public), “true AI” involves unsupervised machine learning (ML) — a computer, software and algorithms that can “think” on their own. AI’s less sophisticated (but much more well-known) cousin machine learning needs supervision to learn, though the technology is certainly capable of high-level fraud prevention and business operational tasks — which is why financial institutions have adopted ML, as the report has shown.
According to the PYMNTS research, data mining stands as one of the most popular algorithmic technologies among U.S. financial institutions — 200 of which, ranging in size from $1 billion to more than $100 billion in assets, provided information that contributed to the findings of the AI report. The research found that nearly 71 percent of banks reported using data mining, defined as a statistical method to extract trends and other relationships from large databases.
Data mining is not true AI (more about that in just a bit), but how it is used illustrates another important trend involving AI and ML: the correlation between a bank’s size and the sophistication of its learning systems, with larger banks typically using more sophisticated systems than smaller ones. When it comes to data mining, for instance, 95 percent of large banks and 79 percent of mid-sized banks use it, the report found. Meanwhile, just 61 percent of small banks reported using data mining technology — a majority, but not nearly as prevalent as it is among larger FIs.
True AI systems, by contrast, are used by only 5.5 percent of financial institutions, as their interviews were used to help construct the report’s findings. Far more popular — besides data mining — were less sophisticated technologies, including BRMS, which enables companies to easily define, deploy, monitor and maintain new regulations, procedures, policies, market opportunities and workflows.
Large Versus Small Banks
In the case of BRMS, use has dropped off significantly for the largest banks.
BRMS’ were used by 77 percent of mid-sized banks, 84 percent of large banks and only 55 percent of the largest banks. That trend also helped for the next most popular algorithmic tech behind BRMS: case-based reasoning (CBR), defined as an algorithmic approach that uses the outcomes from past experiences as input to solve new problems. The two groups of banks that were most likely to use CBR were mid- and large-sized banks, while just 18 percent of the largest banks in the sample reported using it. Even small banks were more likely to use CBR at 26 percent.
Larger banks are more often drawn to advanced machine learning technologies. That includes fuzzy logic, used by 14.5 percent of the banks in the PYMNTS report, and used almost exclusively by the largest banks. While traditional logic typically categorizes information into binary patterns like black/white, yes/no or true/false, fuzzy logic presents a middle ground where statements can be partially true and partially false, accounting for much of humans’ day-to-day reasoning.
In addition, 8.5 percent of banks in the report used deep learning and neural networks, technology loosely inspired by the structure of the brain, with a set of algorithms that use a neural network as their underlying architecture. Banks that do use deep learning, however, tend to be among the largest: 91 percent of the largest banks reported using it.
More generally, 100 percent of all financial institutions in the report said they use at least one form of machine learning technology. One can think of machine learning, maybe, as training wheels for AI.
However, don’t take that too far. As the report noted, “use of the term ‘AI’ has not only created confusion, but it has diluted the power and the impact of this incredibly powerful technology on payments and financial services.” Such misunderstanding can lead FIs down a wrong path, because they have invested billions of dollars in legacy approaches that are largely manual and repetitive.
As the report noted, “This includes consultant fees, armies of back-office agents and outdated rules that flag violations of [anti-money laundering (AML)] regulations, which they describe as AI. These systems have proven to be largely ineffective at actually curtailing money laundering and, as a result, regulators in the United States and the European Union have issued more than $340 billion in fines for non-AML compliance since 2009.”
When it comes to artificial intelligence systems (true AI), the report said that common uses are enhancing the consumer experience and fighting fraud. PYMNTS found that 81.8 percent of financial institutions use AI for banking services, and 72.7 percent use it to fight internal fraud.
“AI systems function similarly to deep learning systems, gathering and storing data that will be used to execute more complex, calculated functions later on,” the report noted.
When Playing By The Rules Doesn’t Work
The report also found evidence that using supervised learning instead of true AI can lead to operational shortcomings. Take those BRMS’, for instance: 37 percent of respondents said the tool was problematic because it often required manual intervention, but about 50 percent listed that as a benefit.
“In simple terms, BRMS’ are automated; this is both their strength and their weakness. It is a strength in the sense that it cuts the cost of operations, but a weakness because, in practice, many companies often encounter situations where they must make exceptions to their usual operations to optimize business,” the report said.
That is one of the major shortcomings of these rules-based systems: It’s hard to make rules that keep pace with the dynamic nature and behavior of consumers, as well as the dynamic nature of the cybercrooks who seem to know the rules as well as FIs do — and play by them so well that their bad deeds escape detection.
While true AI is still a tiny part of the banking world, the report argued that all FIs could benefit from “smart agents,” an AI application.
“Smart agent technology is a personalization technology that creates a virtual representation of every entity it interacts with — including customers, banks and others — and learns by building a profile from that entity’s actions and activities,” the PYMNTS report said. “Smart agents are also highly adaptable and can be used in a wide variety of contexts to enhance customer-facing operations and services. In the payments sector, smart agents gather and store online information about customers, point-of-sale (POS) terminals, merchants and other entities, using it to personalize the services they provide.”
Another benefit (one that puts smart agents into more practical terms, perhaps?) is that there can be as many smart agents as active entities. That means 200 million smart agents could be analyzing 200 million cards engaged in transactions.
Though no financial institutions that spoke with PYMNTS have yet adopted smart agents, interest is high, especially among larger banks (that is, banks with more than $100 billion worth of assets). Of those banks, nearly 73 percent said they were either “very” or “extremely” interested in smart agents.
Such a finding argues that AI, still in its infancy, will find wide acceptance over the long term in the payments world. The numbers might be low now, but there is much room — and so many reasons — for them to grow.