The future of real-time payment starts with AI

“AI has the power to rapidly synthesize a multitude of information from a wide range of data sets and provide solutions accordingly. It can handle steps in processes without the need for human intervention, enabling faster payments.”

Markus Veith

Grant Thornton National Leader, Blockchain, Digital Assets and Web3 Solutions

“AI has the power to rapidly synthesize a multitude of information from a wide range of data sets and provide solutions accordingly,” said Grant Thornton Blockchain, Digital Assets and Web3 Solutions National Leader Markus Veith. “It can handle steps in processes without the need for human intervention, enabling faster payments.”

But beyond real-time payments, AI can play a role in a wide range of banking operations, creating efficiencies for both customers and operations.

Improving the customer experience

The use cases for AI in the banking industry are almost endless. One is to help improve customer satisfaction and create new opportunities for customers. For example, Tasman suggested that AI tools can make predictions and recommendations based on past consumer behavior, finding ways to tap into a customer’s buying habits and using that to meet customer needs.

“Maybe patterns of spending in different parts of the year, such as the holiday season, tick up and there’s a delay in paying off a credit card in the new year,” he said. “AI might be able to predict that type of behavior in the future and offer different types of payment options ahead of those purchases in future cycles.”’

Tasman added that the ability of AI to track past or recurring customer behavior also opens the door for banks to offer them new products. “There is an opportunity to offer the customer a choice or an opportunity to invest in different products based on their prior behaviors,” he said.

Assisting in internal processes

And beyond transaction data, AI can improve tools like chatbots and automated voicemails to provide more intelligent, curated responses for a personalized customer experience on a 24/7 basis. Grant Thornton Internal Audit Cybersecurity Practice Managing Director Vikrant Rai said these upgraded tools can diminish the need for call center staffing, adding to banks’ cost savings and improving efficiency internally and for clients.

“AI-powered software can be used to create intelligent chatbots that instantly address customer inquiries and in some advance-use cases, even use camera scans to know if a customer is happy or satisfied,” Rai said.

Furthermore, AI can improve the lending process by cutting down on underwriting time, deepening customer insights, including credit scoring, creating personalized training for employees, assisting banks in examining their positions more closely to identify superior investment opportunities and much more.

Mitigating risk

Another opportunity with AI in banking? Risk management. Yes, really.

“As technology gets smarter, it’s important to remember that risks get bigger, too,” Veith said. “AI and machine learning can support risk management in a variety of ways. From scenario planning to asset-liability committee operations, AI tools can play a role in risk management processes by capturing patterns, noting anomalies and mitigating threats.”

But despite these use cases, AI and machine learning (ML) techniques have risks of their own. Before banks can begin incorporating these tools into their processes, they should consider the inherent risks and put the right safeguards in place.

Build data safeguards into your strategy

The risks of incorporating AI are present in all industries but are of particular concern to banks and related payment partners. Fraud, data privacy failures and the use of inaccurate data or models can cause significant damage. These risks are intensified because data is often being swapped between banks and third-party vendors.

And those risks are possible even when an AI model is well-designed and trained with consistent, reliable, relevant and timely data. But when open source tools are used, it can be difficult to ascertain the quality of the model and the data upon which it relies.

"Banks should determine what data they want to collect and from what sources. AI provides the ability to rapidly source across a multitude of information or datasets. Part of your strategy should include what datasets you want to pull from — both traditionally and with future capabilities with the power of AI — and how those could work together.”

Scott Tripp

Grant Thornton Managing Director, CFO Advisory

“The results of generative AI are as good as the data that goes into it,” said Grant Thornton IA, IT and Cybersecurity Practice Leader Scott Peyton. “Say you use a generative AI model that’s trained on every detail about Earth. If you ask it about those twinkling lights in the sky, it will never know the answer is ‘stars.’”

To incorporate AI tools into banking operations and real-time payment processes, banks need to be working with a mature, reliable model. So the first step is building an AI strategy. An organization-wide AI strategy includes the critical element of board oversight, which verifies that processes are in place to keep AI use consistent with the financial institution’s mission and principles.

Grant Thornton’s CFO survey for the third quarter of 2023 showed that just 48% of CFOs said their boards have taken an active role in understanding governance over generative AI. Bank boards can and should do better.

A key governance task is verifying that management has implemented an AI risk management framework that’s appropriate for the organization.

Although many different frameworks exist, one that many organizations are finding useful is the National Institute of Standards and Technology’s AI Risk Management Framework, which focuses on managing risks to individuals, organizations and society associated with AI.

Once a direction on AI is established and a framework is in place, banks and related partners in the payment spaces need to address the building blocks of AI by developing and sourcing the right data. Scott Tripp said, “Banks should determine what data they want to collect and from what sources. AI provides the ability to rapidly source across a multitude of information or datasets. Part of your strategy should include what datasets you want to pull from — both traditionally and with future capabilities with the power of AI — and how those could work together.”

Rai said data is one of the most essential components of a good AI model.

“Data quality needs to be high and defining that data, controlling who has access to that data, as well as where it’s hosted, is imperative,” he said.

Companies will need to protect data to maintain accurate information about their customer. “Organizations should continue to apply data integrity and data protection controls, as well as necessary data validation controls, to protect against inaccurate bias on data sets,” Rai added.

“It’s also being able to train and correct an AI model,” Peyton added. “It’s important to have the right data inputs, true, but it’s the corrective action that humans need to take that says, ‘you need to respond this way instead’ that make a model reliable and usable in a business setting.”

Testing and validating the outputs of AI still is very much a human endeavor. Although AI can produce results, people need to determine whether those results are reliable.

That’s where the use of AI can get difficult and requires tough questions and decisions.

“First, can we validate the training of the AI model, yes or no?” Peyton said. “And then, can we empirically validate the data and the results the AI produces? If the answer is no, the risk of relying on AI or having GenAI be an influence or an input into our decisions goes up.”

As with any risk, internal audit will play a critical role in keeping an organization’s AI use consistent with its principles and compliant with regulation. But it’s important to remember that AI is an emerging technology for internal auditors as well.

They need a full understanding of the technology to skillfully fulfill their important role. They also need regular updates — along with the rest of the organization — on AI regulatory developments as new compliance requirements emerge.

“Internal audit needs to be trained,” Veith said. “They need to be able to test and audit to prepare for regulators.”

The importance of human intervention

AI tools can rapidly speed up processes in real-time payments and other banking operations, but the technology still has limits.

“When it comes to real-time payments, AI should never be doing everything,” Tasman said. “Its job is to handle things that might become arduous for an individual to manage, like verifying document completion, for example. But many steps in the process need to be properly adjudicated by a trained professional.”

Human intervention is particularly important in the role of detecting fraud, Scott Tripp added. “Algorithms can be easily gamed. So all processes need to have at least some human intervention,” he said.