Why Incremental Progress on AI in Banking Will Be the Best Way Forward
Excitement around the transformative nature of AI doesn't mean it will happen in wholesale. Given the tightly regulated and inherently risky nature of the banking industry, financial institutions should focus on finding and perfecting incremental use cases for AI. But "incremental" will not necessarily be synonymous with "slow."
By Chris Griffin, Co-Founder of Narmi
Simple Subscribe
Subscribe Now!
Federal Reserve Governor Michael S. Barr, speaking recently to a consortium that included bank and fintech leaders, outlined how the Fed sees the role of Generative AI within the future of banking: “an incremental scenario where the technology primarily augments what humans do today, and a transformative scenario where we extend human capabilities with far-reaching consequences.”
Generative AI certainly does have far-reaching consequences. The first wave is providing foundational tools like Microsoft Copilot to bank employees to improve their day-to-day. However, the path to true revolution will be a long road, marked with many narrow applications of AI rather than one sudden, wholesale change.
Incremental change might not sound very interesting. But critically, the rate of change is increasing at a dramatic pace.
Want to read more like this? Check out Narmi’s content portal on The Financial Brand: Be Where Banking is Going
Banking Was One of First Movers in AI
Much of what we refer to as “artificial intelligence” today is what people in the field traditionally called machine learning – technology that has been used in financial services for some time. Machine learning had its roots in the 1970s, but research from S&P Global notes that “[t]he application of machine learning in banking accelerated in the late 2000s… Machine learning in banking, financial services, and insurance accounted for about 18% of the total market, as measured by end-users, at the end of 2022.”
Of course, 2022 was the year that ChatGPT – widely understood as the “killer app” that ushered us into our current fascination with AI – was released. Leveraged for data classification and process automation, machine learning has long been a helpful tool augmenting the back end of banking data flows; generating insights, uncovering hidden patterns, and predicting events in areas like fraud prevention. Now, the conversation in the industry is focused on the shift to Generative AI, which requires less specific training and development, and can frequently deliver better results.
As recently as three years ago there was a wide gulf between the theoretical reliability of LLM-powered AI and its actual use cases. That’s certainly not the case today. The gap has narrowed significantly, and there’s potential for a robust pipeline of AI tools that balance minimal human intervention with significant human oversight. Today, it’s useful to think of the “AI revolution” as the end point of a rich, dynamic process that will realistically take years, a process that all of the stakeholders in our heavily regulated industry – banks (central and commercial), financial service technology providers, and regulators – will work through together.
Key will be stronger partnerships between banks and their most important technology partners, what Governor Barr called a “symbiotic” relationship in which the former’s deep customer data is put to use by the latter, “forming collaborative partnerships where fintechs and banks merge their strengths.”
Let’s use the state of data governance as a case study of where the industry is on its AI journey. According to a recent Cornerstone Advisors report, while 74% of banks and 77% of credit unions would assess the effectiveness of their organization’s data governance as at least “somewhat effective,” that number drops precipitously when asked about their actual use of that data to enhance operational efficiency – 51% and 59%, respectively.
These findings bear out regardless of what survey you’re looking at. Salesforce found that 93% of financial services decision makers agree that their organizations should be getting more value out of their data, and a wide-ranging Gartner survey saw that 48% of financial institutions struggle with a lack of clarity about AI’s business impacts. Thirty-seven percent of Gartner’s respondents described a “production-first mentality,” where trust and risk management are an afterthought, as a challenge to implementing AI governance.
To put it bluntly: however confident financial institutions are in the potential value of their data, many are still at a loss in terms of its application. It’s the banks and financial technology service providers who can deploy narrow, compliant, and useful applications of AI that will win. For banks, this means relying on partners who can provide AI in the right context for them, and who build on processes designed with their unique problems and policies in mind.
Near-Term Problems: Regulation, Compliance, and (Non)Templatization
It should come as no surprise that banking would like to see AI as a magic pill: a recent McKinsey study claimed that “since 2010, productivity at U.S. banks has been falling 0.3 percent a year, on average, even as most other sectors have experienced productivity gains.” The initial use cases of workplace AI are meant to address the areas where much of this lag occurs: Salesforce found that “customer service representatives only spend 39% of their time working with customers, with the rest spent largely on low-value administrative work”.
But Barr’s remarks make it clear the Fed and other regulators will demand a voice in the AI conversation, too. He lauds banks for being “appropriately cautious in the highly regulated environment in which they operate,” and makes repeated references to “responsible innovation” amid concerns about information security and poorly optimized data/business practices. It is likely regulators will make quick examples of the financial institutions that take a reckless approach.
Barr also referenced the stochastic processes inherent to Generative AI. Because Generative AI is nondeterministic by definition, its models will generate a fresh response to every query, even when asked the same thing. In banking, where decisions must be precise and replicable, this lack of templatization greatly increases the likelihood of unacceptable results, like biased credit decisions.
This issue is a fundamental deficiency of Generative AI, and one that is expected to see only incremental improvements. An Apple report from 2024 found that “LLMs struggle even when provided with multiple examples of the same question or examples containing similar irrelevant information. This suggests deeper issues in their reasoning processes that cannot be easily mitigated through few-shot learning or fine-tuning.” Bottom line: It will be important for there to be strong human oversight for some time.
Dig deeper:
- Keeping Payments at Mastercard on AI’s Leading Edge — without Falling Off
- The 5 Use Cases for Adding Generative AI to Your CX Strategy
- An AI System Built for Everything is an AI Built for Nothing
What Does ‘Efficiency’ Mean Now?
Of course, no bank will – or should – be fully reliant on AI. Given that only 10% of consumers fully trust use of AI agents in financial services, widespread implementation of AI workflows and interfaces will be a hard sell until users have grown accustomed to their value. After integrating AI into customer support interfaces like chatbots, banks will likely continue by first enhancing existing machine learning applications with Generative AI, like leveraging those chat logs’ “digital data trails” to calibrate its internal understanding and scale practices like customization.
Says S&P Global: “Notable changes due to the application of Generative AI in banking are unlikely to be immediate… [t]he bulk of banks’ near-term use cases will likely focus on offering incremental innovation… based on specific business needs. Finally, we expect employees will remain in an oversight role, known as human-in-the-loop (HITL), to ensure results meet expectations (in terms of accuracy, precision, and compliance) as the technology matures.”
A financial institution’s employees are the most important variable when considering an AI strategy. Usage of AI technology scales through a bank’s rank-and-file, and efficiency gains are only unlocked through them. But many visions of banking’s AI future seem to put employees last, thinking of them as mere vectors for efficiency and over-anticipating the technology’s ability to replicate their functions.
In our incremental AI landscape, choosing partners that value a financial institution’s relationship with its employees (and those employees’ relationships to customers) will be critical. Humans will play a significant part in banking processes for a long time – as they should, given the risks of getting things wrong in our industry. They should feel empowered by technology, becoming superusers while managing the risk involved with AI processes and ensuring there is no break in the customer experience. Banks have been experts in risk management for a long time – this is their time to shine.
Technology developed under this “staff-to-strategist” model will be necessarily incremental in its approach, taking into account the pace of regulation in a deliberate paradigm shift to AI integration. The companies developing these applications should endeavor to provide clarity and proper deployment controls to the banks using this technology, given the delicate (and foundational) nature of relationships they have with staff and customers alike.
A phased implementation of AI technology allows financial institutions to properly align to business cases, intentionally grow usage to match the pace of development and regulation, and allows staff to grow accustomed to having conversations about AI on terms informed by their experiences. In the near-to-mid term, industry players are best served by this incremental approach by identifying narrow use cases for AI, hyperfocusing on deploying solutions, and consistently checking to see if they are yielding the intended benefits – without introducing unintended consequences.
Given the state of sustained market whiplash we’re in right now, it’s ever-more important that AI in banking doesn’t become the Wild West. The incremental approach is more natural to the motions of the banking industry than brute force.
It’s best said by Governor Barr: “Successful integration of Gen AI into banking will require both creativity in adoption as well as getting the guardrails right. That’s not a zero-sum game.”
