In an era of rapid technological evolution, few industries are as poised for transformation as banking and payments. It is a transformation that’s already underway, with technologies like generative artificial intelligence (GenAI).
“Governance, cost and utility form the three points of the [GenAI] triangle we’re trying to balance. Getting it right will unlock transformative possibilities for our industry,” Mark Sundt, CTO at Stax Payments, said during a conversation for the PYMNTS Series “What’s Next in Payments: Memo To The GenAI Companies.”
As banking and payments companies consider the technology, the focus is shifting toward practical applications of AI that can drive efficiencies, improve customer experiences and address fraud.
With AI tools embedded in nearly every software service, companies must also prioritize guardrails to ensure data privacy and security.
While AI’s potential is undeniable, Sundt highlighted critical concerns around safety and cost. “Governance has been our first priority,” he said. “Every service we use has some AI component embedded in it. Governance is about ensuring that these tools don’t inadvertently proliferate sensitive data.”
Challenges to AI Payment Applications
Cost considerations around AI usage are equally pressing. Large language models (LLMs), while powerful, can be prohibitively expensive to operate, and their utility doesn’t always justify the investment.
But as the industry shifts away from monolithic models that attempt to solve every problem, smaller, specialized “agentic models” are gaining traction. These models are designed to address specific issues efficiently and cost-effectively, marking a shift in how AI is deployed.
“We’re moving away from the kitchen-sink approach of large models to focused, orchestrated agents that solve specific problems effectively and at lower costs,” Sundt said.
This trend, he added, is particularly evident in customer service, where AI is being used to diagnose and resolve issues with precision. Instead of building massive models that attempt to answer every conceivable question, firms are developing smaller models focused on solving tangible problems.
“Customers don’t want encyclopedic answers. They want to know how to configure their terminal to optimize interchange fees and response times. That’s where smaller, specialized models shine,” Sundt said.
“This pivot toward dynamic, smaller models is akin to the evolution of computing itself,” he added, noting that it parallels the broader transition in technology from mainframe computing to distributed cloud applications.
AI Fraud Detection, Risk Management
AI’s role in risk management and fraud detection is another area of immense potential.
“The biggest red flags we encounter are merchants with newly established banking relationships or websites. These temporal attributes often signal fraudulent intent,” he said. Stax leverages AI to enrich decision-making by evaluating factors like account longevity, storefront existence and transactional behavior.
Sundt also described suspicious patterns in transactional fraud, such as large transactions followed by batch reversals or refunds issued to different credit cards. “These scenarios demand robust AI systems to detect and mitigate fraudulent activities at scale,” he said.
Looking ahead, Sundt believes the next frontier for AI lies in reasoning.
“Current models excel at summarization and categorization, but they struggle with reasoning. Humans make decisions with incomplete information all the time. Teaching AI to do the same would be a monumental leap,” he said.
This capability would enhance everything from fraud detection to customer service, enabling AI to navigate complex, ambiguous scenarios with greater sophistication.
Still, Sundt underscored the need for AI providers to align their rapid development cycles with the slower, compliance-driven nature of the banking industry.
“The financial sector doesn’t move at the speed of tech. It’s crucial for AI companies to understand our industry’s unique challenges and design solutions accordingly,” he said.
