Preparing for agentic AI in payments

Imagine having a financial assistant pay your bills on time, alert you of any suspicious transactions and rebalance your investment portfolio based on real-time market data. That assistant could handle the accounts payable function for your business, including verifying invoices and purchase orders with delivery confirmations and automatically initiating payments.

That “assistant” is agentic AI.

AI agents can be thought of as assistants, though they can be proactive and autonomous. They have the ability to understand, plan and complete actions to achieve targeted goals.

With banking, they can manage complex, multi-step financial workflows and tasks — such as credit underwriting and fraud detection — and learn from interactions with minimal human input to improve efficiencies and enhance customer experiences. For example, an AI agent could alert you to a suspicious transaction then complete a funds transfer upon a verbal approval.

With agentic AI payments, financial transactions are done autonomously by AI agents, software that can perform an action based on instructions using context on behalf of businesses, devices and users. Agentic payments empower AI to make real-time decisions on if, when and how money should be moved, based on pre-defined rules, machine learning and user intent.

For banks already using the technology, results show up to a 40% reduction in cost and 30% increase in revenue growth.

The agentic e-commerce opportunity

A primary focus and use of agentic AI is on the digital marketplace. Known as agentic commerce, online shopping is powered by AI agents acting on one’s behalf.

The technology anticipates a customer’s needs to search and compare products, negotiate deals and complete transactions. All of which would be done with a person’s intent.

Are consumers in favor of machines making their purchases?

According to one study, agentic AI is set to influence more than $1 trillion in e-commerce spending. Additionally, 81% of U.S. consumers expect to use agentic AI tools to shop, which will influence more than half of all online purchases in the near future.

A stark advantage of the technology is that agents can operate across multiple vendors, tabs and merchants, evaluating options based on a person’s preferences, then completing a transaction.

The process of bringing AI agents to life is already well underway, said Jack Hilger, Senior Director of North America Product, Visa.

“You can imagine a world where sophisticated search looks like an agent inquiring about availability at a new restaurant in town and then booking a reservation that works best with your schedule. Or maybe it’s booking a complete vacation — end to end — optimized to your preferences and working entirely within your budget,” he said. “Sounds like sci-fi, but I promise we’re not that far away.”

Risk and oversight

With its vast potential, major banks are already starting to embed agentic AI into their offerings.

Like any up-and-coming technology, however, it’s not without risk and uncertainty. Some of the primary challenges include:

• Security and privacy: Eight in 10 highly automated firms cite data security and privacy as their top concerns. That’s more than double the 39% reported by less-automated firms. When AI agents interact with more sensitive information, the risk increases. Banks should assess their security frameworks and how they monitor potential emerging threats.
• Governance and regulatory: The legal and regulatory landscape for AI agents is evolving. Without clear guidance, companies risk falling out of compliance with regulations such as the General Data Protection Regulation (GDPR) and the EU AI Act. This could result in fines and legal ramifications.
• Operational, integration and implementation: Major U.S. firms across the goods, technology, and services sectors often cite the integration and implementation of AI as a significant operational challenge.
• Resistance to change: Just 24% of consumers are comfortable letting AI complete a purchase on their own. Add in potential fears of job displacement — or a preference for manual processes — and it’s easy to see how mass adoption could take some time. Banks could offer training and clear communication about the benefits to help ease some of the tension.
• Managing systematic bias: If AI systems are trained on data that reflect biases, the technology can perpetuate and amplify those biases. This could lead to discriminatory decisions. Banks should monitor and audit their AI models to prevent biased outcomes.

It all starts with data

Garbage in, garbage out.

For agentic AI, it’s data in, data out when it comes to ensuring more favorable outcomes from AI systems. Monitoring the information that’s going in and out of the models helps ensure the effectiveness of the agents.

They rely on clean, complete and high-quality data.

Since agents operate autonomously, errors or biases in the data could be amplified. This could lead to inaccurate outcomes.

If you are feeding an AI system with bad data, you are going to get bad outcomes, said Ken Viciana, Head of Data and Analytics, TSYS.

The trust factor: Can humans and AI work together?

By giving automated AI agents the power to make decisions, could companies create a gap between humans and machines?

Take the case of Swedish fintech provider Klarna. The company launched an AI assistant it claimed could do the work of 700 full-time contact center agents.

After a year, however, there was a drop in customer service and experience standards, according to Bloomberg. Klarna then pivoted, turning back to people to handle more customer service work so customers had the option to speak with a live person.

The lesson was that AI does well when you augment it with humans and vice versa, said Arun Ramanathan, Senior Gen AI Strategist, Amazon Web Services.

“Oftentimes we’re so involved at using technology to transform processes and improve efficiencies in our daily lives that we do not think about human interaction,” he said. “How can banks boost efficiency without sacrificing that human connection?”

Finding a middle ground that focuses on the customer.

“In payments, you must have a very human-centric and strategic ethical deployment to make sure it’s successful,” Ramanathan said. “You can be not only efficient, but also empathetic. That’s going to be key.”