Five Real-World AI Applications That Will Boost Your Bank’s Operations
By Steve Cocheo, Senior Executive Editor at The Financial Brand
Simple Subscribe
Subscribe Now!
Executive Summary
- Artificial intelligence is often spoken of as a monolithic entity. But experts at Cornerstone Advisors say practical application of AI techniques hinges on breaking it down into multiple tools.
- Enterprise applications of AI — not just GenAI — are producing improvements in lending, payments, risk management and more.
- Key to realizing the gains: Stop putting AI in one team’s hands. It should be woven throughout the bank.
Banks are famous for “silos,” organizing operations into rigid sub-organizations that don’t communicate with each other. Is the industry repeating that pattern with artificial intelligence?
“Banks and credit unions have to stop thinking about AI as a function. They can’t be thinking about AI in terms of a ‘Chief AI Officer’ or an ‘AI initiative’ or an ‘AI policy’,” said Tony DeSanctis, senior director at Cornerstone Advisors, during a recent webinar. “We need to be thinking about AI being integrated into everything we do.”
Infusing AI into all parts of the organization is essential to realizing the technology’s potential, DeSanctis believes.
More important, Ron Shevlin, managing director and chief research officer, advised webinar listeners to stop using AI as a blanket term. Distinct flavors of AI will drive the applications that can help banks and credit unions reimagine their operations.
“Talk about conversational AI, generative AI, machine learning, agentic AI, and other types,” said Shevlin.
DeSanctis and four other subject matter experts from Cornerstone pinpointed five areas where specific, enterprise-level uses of AI can improve efficiencies and results.
Want more insights like this? Check out Candescent’s content portal: Illuminating Insights in Digital-First Banking
1. Lending: Finding the Human Lender/AI Balance Point
Use of artificial intelligence decisioning has already had time to prove itself, and the results have been strong, according to Daryl Jones, senior director. The fit varies from one institution to another, “but the lift, overall, has been unquestionable,” said Jones. He said institutions using AI in lending decisions have generally seen healthy increases in approvals, with solid results.
One caveat is that as aspects of loan decisions transition to AI, institutions have to be careful how human lenders influence the software development process. Jones said that sometimes lenders attempt to bring elements of legacy systems into the new AI models because they think they will perform better. “Generally, that creates issues,” he warned, and drags down performance.
DeSanctis also warned institutions to be careful what data they train loan decisioning AI on. He noted that one company he knew of trained an AI resume screening tool on the CVs of their top 100 software engineers. This introduced a bias towards male applicants, and the software filtered out resumes from graduates of notable schools with a history as women’s colleges.
“AI has zero conscience and zero ethics,” said DeSanctis. Banks, subject to fair-lending and other compliance rules, have to anticipate potential problems, he advised.
Prospecting is another task that AI can help with, according to Jones. Combinations of AI types can, for example, evaluate a mortgage portfolio to see where opportunities to offer refinancing exist. Then the systems can initiate contact with potential refinance borrowers, identify those who are interested, gather preliminary information, and set up appointments with human loan originators.
Jones also said that an early but growing application is higher-level chatbots that handle taking applications from prospects, posing questions and recording answers, and assembling the data into application-friendly form.
“This approach is going to drive better engagement, better borrower satisfaction, and higher application completion rates,” according to Jones.
Read more: Can AI Really Deliver Savings and Efficiency? What Big Banks Have Learned
The Other Side of Lending: Collections
Types of AI are helping with the other end of the lending process — collections — as well. Jones said that AI analysis of borrower behavior can help improve collection efforts down to the level of the individual borrower.
“When a borrower’s called, what time of day are they picking up? Are they reading texts? Are they reading emails? Do they respond better to texts than to phone calls?” asked Jones. Evaluating past patterns can guide more effective collection strategies than merely having collectors continually working the phone.
Similarly, early-stage collection efforts using outbound AI collector application are showing promise.
2. Customer Contact: Creating Smoother Communications
Among Cornerstone clients, the median time to handle inbound customer calls is six and a half minutes, according to Ryan Brogan, managing director. Automating more of these interactions via AI is one of the most effective AI applications in customer service, he said. He said using “agent assist” — a combination of natural language processing, machine learning and retrieval augmented generation, institutions can handle in seconds some queries that ordinarily take minutes.
Brogan warned that some of the resources that human representatives routinely tap aren’t readable by AI — embedded videos, for example, in bank databases — so some restructuring or revising will be necessary.
Making sure callers are who they say they are can be a challenge.
“The way that banks have historically verified customers in banking is not great,” said Brogan. The typical knowledge-based questions “can often feel like an interrogation,” he said, and contact staff frequently feel the pressure of being the only line of defense.
Brogan said that voice biometrics technology can learn the cadence, style and tone of an individual’s voice — recording and checking it later by using machine learning and neural networks. It can reduce verification to seconds and is becoming more available and cheaper.
Still, AI-based deepfakes of video and audio have been getting more and more realistic. Brogan doesn’t think voice biometrics will be sufficiently secure on its own, “but it’ll be part of a layered approach.”
Read more: The AI Advantage: How to Build a Future-Ready Workforce with Smarter Training
3. Risk Management: In the Moment Instead of in the Rear-View Mirror
Technology has long been a mainstay for antifraud, according to John Meyer, managing director.
“We’ve had machine learning algorithms since the 1990s,” said Meyer, but today’s antifraud applications of AI go a step beyond. He explained that the old technology could evaluate a few data points “on day two,” once the damage was already done.
By contrast, AI-based techniques can screen and surface instances truly needing human evaluation, according to Meyer. Such applications include verifying that paper checks are genuine. Meyer noted that check fraud remains a significant issue for the banking industry in spite of the rise of digital transactions.
“Fraud is always a math problem,” said Meyer. “Am I actually stopping more stuff [whose value is more] than I’m paying for the fraud system itself?”
AI is helping more institutions with digital banking fraud and compliance issues as well. Increasingly, potential fraud can be flagged and addressed right in the midst of a questionable transaction.
Meyer explained that AI can differentiate behavioral patterns of genuine users versus fraudsters. Take a mobile banking interaction: Many consumers typically go into their account, check their balance and recent history, before doing what they came for.
“They bounce around a bit,” he said, and AI can be trained to spot divergence from established patterns. The latter include keyboard pacing and how the user tends to navigate.
On the other hand, “the bad guys go right into making a transfer,” and AI may spot the different pattern. If the AI indicates potential fraud, suggesting the phone has been stolen, he said it can insert some additional friction on the fly. This might include adding a multi-factor authentication step, requiring a callback.
“This is real-time interdiction as opposed to day two detection,” said Meyer.
Read more: Auto Buying Fraud is Exploding. Capital One Is Using AI to Fight Back
4. Payments: Getting Ready for the Agentic Wallet
Middle-aged DeSanctis joked that younger co-workers kid him for buying things in stores with a physical card, versus purchasing online and on mobile. He said that he’s waiting for the day when still-younger people can razz them for not buying things with an agentic wallet — a wallet that provides transaction capabilities for the use of a consumer’s agent AI, empowered at some level to shop and make purchases on the consumer’s behalf.
Bankers have long talked about being “top of wallet,” the first choice in a transaction, first in physical wallets, then in digital wallets.
“I think being ‘top of the agentic wallet’ is going to become critical,” said DeSanctis. “You’ll want to make sure that your cards get added into it.”
Agentic AI is in its early days for commerce and payments, but banks and credit unions need to monitor this closely. Visa and Mastercard are already building out agentic services, he said.
Read more: Three Must-Dos: Faster Payments, Stablecoins and Agentic Commerce
5. Back Office Operations: AI Has a Home Here, Too
Even in a modern banking office, documents can be a rat’s nest.
“We had a client on the West Coast that wanted to centralize all of its operational documents,” said Clio Silman, managing director. “We’ve all been there, right? They’re in Word, they’re in PowerPoint, they’re on somebody’s desk, they’re on somebody’s hard drive, they’re in different formats.”
Pulling everything together is a mammoth task. “It typically falls to a few people with some knowledge of the documents,” said Silman.
Instead of that, the client rolled out a tool that came equipped with multiple AI capabilities. The AI, with human input, framed a set of procedures for centralizing the documents.
The result paid off immediately — efficiency improved because the paperwork got organized. But it also produced ongoing dividends because centralization clued everyone into how and where to find documentation. As changes to procedures are made, the AI automatically updates its approach.
“It’s the project that keeps on giving,” said Silman.
Read this next: Why AI Won’t Replace Bankers but Will Expose the Bad Ones
