In many ways, artificial intelligence (AI) is all but a settled technology for much of the payments industry. While a year ago many were scrambling to find the best ways to apply the technology, today it has widespread use across many applications.
While a common application for older iterations of AI, fraud and compliance have been stepped up with generative AI (genAI) enriched solutions, with Mastercard and Swift among those to harness the technology in new fraud-fighting initiatives over the past year.
Customer service has also benefited, with Remitly and Klarna among those to drive significant operational efficiencies from customer-facing genAI – the former reported that it had reduced customer support times by 75% using the technology. Meanwhile, Klarna has been a particularly strong adopter of AI, implementing it throughout the company amid efforts to halve its workforce ahead of its upcoming IPO.
However, while the technology has begun to provide measurable operational benefits, the next step beyond genAI is now emerging, in the form of agentic AI.
Agentic AI: Artificial intelligence’s next big opportunity?
In some senses, agentic AI has the potential to deliver much of the promise of artificial intelligence. When genAI often sees a back and forth between user and AI, refining requests and breaking down tasks into measurable units to get the best outcomes, agentic AI is about autonomy.
AI agents are intended to act independently, with the ability to make complex real-time decisions, learning dynamically in the process and interacting with technological systems or even other AIs as required.
As with any technology, the full range of applications is still being realised, but areas with potential application include the management of investment portfolios, proactive AI-based personal assistants and smart virtual healthcare agents. In the business world, there is also potential for a next generation of customer support, as well as applications in DevOps and related infrastructure management.
While we have seen a rise in the number of startups building agentic AI solutions, there are also several major players getting in on the act. In January, OpenAI launched a research preview of Operator, an agentic AI designed to work independently on repetitive online tasks, while Salesforce and Zoom are among those who have also released agentic AI-based features.
What agentic AI could mean for payments
In payments however, the potential of agentic AI is beginning to be felt. At a recent investor day, PayPal highlighted how it was exploring the use of the technology to conduct research and make purchases on behalf of users, while eBay is piloting OpenAI’s Operator to enhance access to site listings.
Stripe has also begun to engage with the technology, launching an optimised checkout solution for AI agents, as well as a solution to enable them to spend money using virtual cards. CEO Patrick Collison and President John Collison have also stressed that there are built-in spending controls, quipping that “nobody wants a T-1000 going wild on the family credit card” in the company’s recent annual letter.
These are early stage moves in the space, but they indicate what could become an immense area for payments: transactions with no human involvement. There are myriad applications for agentic AI that see it initiate and complete transactions on behalf of a user or as part of a wider project, and tools that can effectively and efficiently support the making and taking of these payments are vital.
Such transactions have the potential to further increase the velocity of money, which has already seen significant gains through digitization and the shift to real-time payments, but they will require their own ecosystem of solutions, plug-ins and regulatory oversight.
Agentic AI provides a whole new avenue of potential payments applications, but it will require legislation, security and support to evolve with it. With some players already moving in the space, we are likely to hear a lot more about the technology in the coming months and years.
Read the full article here
