Green Sheet interviews Finastra CPO, Payments, Radha Savanna

AI agents, faster cross-border transfers and emerging payment methods are prompting banks to reassess how their systems support innovation and control. Radha Suvarna, chief product officer, payments, at Finastra, discusses modernizing existing infrastructure, choosing among payment rails and putting AI to work with human oversight and measurable goals.

Green Sheet: 1. How does the rise of AI agents change what banks need from payments infrastructure originally designed for people and institutions?

Radha Suvarna: We believe it is going to be complementary to humans assisting to improve efficiency while keeping the human in the loop. Today's payment infrastructure was built with a human in mind, someone who logs in, reviews, and approves each step. AI agents will likely improve efficiencies in portions of the workflows with a human validating as a checker. Finastra's approach centers on "AI with human in the loop," letting AI handle the heavy lifting while keeping a person validating the moments that matter most.

GS: 2. How far can banks modernize existing payment rails with AI before replacing underlying infrastructure becomes necessary?

RS: We don't believe the underlying payment infrastructure will be replaced by AI; it will be complemented. Layering AI onto the rails banks already have can capture much of the value a full overhaul promises: faster fraud detection, quicker payment repairs, stronger resilience, without the multi-year rebuild.

GS: 3. As cross-border payments approach domestic speeds, what barriers still prevent the two experiences from becoming comparable?

RS: Speed itself isn't the obstacle it used to be. At Sibos, Swift pointed to a payment moving from China to the U.S. in 15 seconds. While that may not be the general average, cross-border payments are starting to become faster and will continue to do so with various cross-border alternatives and initiatives such as Swift Scheme. What is also important is the transparency and visibility: banks and their customers need real-time insight into where funds are and how they can be used, not just confirmation that a payment landed quickly.

GS: 4. What are the less visible costs and risks banks incur by continuing to operate fragmented legacy payment systems?

RS: The visible cost is maintenance: patching and running disconnected systems, disparate operational staff and workflows for distinct rails. The less visible cost is slower innovation, since every new product, compliance requirement or AI capability must be built and tested separately in each system. That drag increasingly shows up as security exposure, too, because every additional integration point is another place where access controls can fail, thus resulting in more vulnerable systems.

GS: 5. As stablecoins, tokenized deposits and traditional rails coexist, how should banks decide which rail best fits a particular payment?

RS: The decision should follow the payment use case and end-customer requirements, not the technology. Speed, finality, cost, settlement speed, liquidity, scheme rules are some of the attributes that will drive the choice of rail. The key is to have the flexibility in an ISO20022 payment hub that enables intelligent routing and orchestration with the choice of best available rail in real time, rather than committing upfront to one. This will provide the foundation for the innovations in emerging use-cases and rails.

GS: 6. What will separate AI payment initiatives that deliver measurable value from pilots that never make it into production?

RS: The initiatives that scale start with a narrow, well-defined problem, like cutting payment repair time or reducing false positives, rather than a broad AI ambition. They also built in human oversight from day one instead of adding it later. The current wave of unauthorized AI agent access across the banking industry is a clear signal that retrofitting controls after deployment costs far more than designing with them from the start.

Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.