AI in payments: building trust, governance, and value across a multi-rail ecosystem

At Sibos, The Paypers spoke with Deepak Gupta, Chief Product Engineering and Delivery Officer at Volante Technologies, about how AI is reshaping payments and what financial institutions need to do to adopt it responsibly.

Drawing on Volante Technologies’ recent report with The Paypers, ‘Payments Reimagined with AI’, Deepak discusses the shift from AI experimentation to production, the importance of trust, governance and human oversight, and how AI could transform payment orchestration across traditional and emerging rails. He also looks ahead to embedded payments, interoperability between traditional rails and digital assets, and the potential impact of increasingly autonomous AI systems.

It’s great to speak with you, Deepak. How are you finding the atmosphere at Sibos so far, and what stood out from your conversations with the payments community?

At the previous Sibos, much of the excitement was around the next payment rail, whether that was real-time payments, FedNow, stablecoins, or tokenised deposits. This time, the conversation has shifted decisively towards AI.

Wherever you go, people are discussing how AI could complement, or even change, the way payment systems are built and operated. More importantly, the conversation has quickly moved from ‘What can AI do for me?’ to ‘How can I use AI to create real business value, and how quickly can I put it into production?’

Normally, new technologies move through a gradual adoption curve, from early adopters and proofs of concept to broader production use. AI seems to be moving much faster. Financial institutions are already asking how it can improve customer experience, generate additional business from existing customers, and provide greater predictability, reliability, and quality of service.

The focus is increasingly on moving AI into production and making it useful in real payment environments.

You work at the intersection of AI and payments at Volante Technologies. What makes applying AI to payments particularly challenging?

One of the fundamental challenges is that AI is inherently probabilistic, while payments require deterministic outcomes. My role is to help bridge that gap and ensure that AI can be used to deliver predictable, controlled results within payment systems.

It is not enough to apply AI to a particular operational task and call that an AI-enabled payment system. The more important questions are what should be delegated to AI, how much should be delegated, and what risks come with that delegation.

Those decisions will also evolve as the technology develops. Something we cannot delegate to AI today may be possible in six months, while other use cases may require much stronger controls.

That is why AI in payments needs to be approached as an evolving roadmap, with clear boundaries around what an AI system can and cannot do.

Your presentation at Sibos connects closely with the findings of Volante Technologies’ recent report with The Paypers, ‘Payments Reimagined with AI’. One of the report’s central themes is trust. What does trust in AI-powered payments mean from your perspective? 

For me, trust comes down to three things: predictability, repeatability, and auditability. Payments also require us to think carefully about the consequences of getting a decision wrong. Some decisions are reversible, while others can have significant financial, reputational, or customer consequences. That means institutions need to establish clear risk thresholds around what an AI system can and cannot do. 

For example, a bank could allow an AI agent to modify a limited number of payments within a defined value threshold. If the agent needs to operate beyond that threshold, additional controls or human approval should be required.

This creates a gradual path towards trust. At Volante, we think about agents operating initially in shadow or learning mode. They can observe behaviour and make recommendations without influencing payment decisions. A human remains in the loop, and the institution can evaluate the agent’s performance before gradually relaxing the restrictions.

What needs to be in place before a bank can confidently allow AI to make payment decisions? 

First, banks need a payment system that can apply AI across different types of payments, even if they initially use it for only one payment type. For example, a bank could start with instant payments and limit the value of transactions on which an AI agent can operate. The agent could first run in shadow mode, allowing the bank to observe its behaviour. From there, the institution can progress through training, proof of concept, and controlled production before expanding its use. The important point is that AI should not be treated as a standalone capability for one payment type. Banks need to think about how they will apply AI consistently across their payment infrastructure. That makes a multi-rail platform increasingly important. It should support traditional rails such as Fedwire, RTP, FedNow and ACH, while also being able to accommodate emerging rails such as stablecoins and tokenised deposits.

With a common platform, banks can establish consistent rules, guardrails, human controls and mechanisms for stopping or limiting AI-driven activity across payment types.

Ultimately, AI will only succeed if it creates tangible value. That could mean more predictable settlement times, fewer payment rejections and returns, or improvements in service levels. AI cannot simply be adopted because it is new or technically impressive. It has to solve real business problems for financial institutions and their customers.

How can financial institutions balance the speed and intelligence AI offers with security, governance and accountability?

It comes down to people, process, and technology. It is not just a technology question. Financial institutions need a platform that is secure, reliable, auditable, scalable, and predictable. Around that technology, they need clearly defined processes, controls, and guardrails, with humans remaining involved in the decision-making process, particularly in the early stages.

Over time, human involvement can decrease as confidence in the models increases, but the agents must remain controllable and auditable. It is also important that every AI-driven decision comes with an explanation. Even if the decision is correct, it is not sufficient if the institution cannot understand why it was made. In payments, context matters.

This is where the distinction between probabilistic AI and deterministic payment outcomes becomes important. The technology provider needs to be able to translate the probabilistic nature of AI into controlled and predictable outputs.

That requires multiple levels of testing, an ongoing review of processes, and a willingness to adapt as the technology evolves. Governance cannot be something that is designed once and then left unchanged.

There is also a cultural dimension. Financial institutions need to create an environment where controlled experimentation is possible. If organisations become so risk-averse that employees cannot experiment, innovation can be stifled very quickly.

You mentioned AI agents in your presentation. How do you see agents being used within payment operations?

The real value of AI will not come simply from applying it to individual operational decisions, such as identifying potentially fraudulent transactions. That is only one part of the payment flow. The bigger questions are what should be delegated to AI, how much should be delegated, and what the risk of that delegation is. Those decisions will also evolve as the technology develops.

At Volante, we are looking at agents that can perform different roles. For example, a predictive agent could identify potential issues before they occur, while a repair agent could identify a potential problem with a payment and recommend how it could be corrected.

Imagine an account number appears inconsistent with the beneficiary name. An AI system could identify several possible alternatives, assign confidence levels to each, and present those options to a human. The human makes the final selection, and the system can learn from that decision.

This is another reason why human-in-the-loop models are important. Banks can allow AI agents to operate in shadow mode, observe their behaviour, and learn from the decisions made by people before allowing the agents to influence live payment decisions.

The objective is not to remove humans from the process immediately. It is to create a controlled path through which institutions can build confidence in AI.

Many banks are still working through legacy modernisation. Why should they be thinking about AI now rather than treating it as a future upgrade?

Different banks are at different stages. Many are still working to move away from mainframe-based systems and towards modern architectures. The risk is that a bank chooses a modern platform today without considering its AI capabilities, and then has to undertake another major upgrade a few years later to introduce AI.

Payment modernisation is already expensive, time-consuming, and risky. Banks should therefore choose a platform today that can support where payments are going tomorrow.

AI should be part of the decision-making process from the beginning. The right partner should be able to support the payment capabilities a bank needs today while providing a path towards AI-enabled operations and multiple payment rails. Rather than having one platform for ACH, another for wires, and separate systems for emerging payment types, banks should consider platforms that can support multiple rails with a common AI layer, common controls, and a long-term vision for how AI will evolve.

Looking further ahead, how do you see AI changing the way payments themselves are experienced by businesses and their customers?

I believe payments will increasingly become invisible. A useful analogy is FedEx. When you send a package, you are asked where it needs to go, when it needs to arrive, and how much you are willing to pay. You are not asked whether the package should travel by air or by another method. The system makes that decision for you. Payments could work in a similar way. A customer could specify when they want the payment to arrive and what they are willing to pay. AI could then determine how the payment should be cleared and settled, which rail should be used and where it should ultimately settle.

The system could also keep track of the value it created, showing the customer how it optimised the payment. In that model, payment becomes less of a product that customers actively manage and more of an embedded capability within a broader financial experience. Banks could also generate more value from services built around payments, rather than from the payment transaction itself.

If we meet again at Sibos in three years, what do you expect the industry to be talking about?

I’m not very good at crystal-ball questions because the industry can change so quickly. A year ago, it would have been difficult to predict how dominant AI would be in today’s conversations.
But I see several areas developing.

First, we will see AI moving beyond operational efficiency, compliance, and fraud detection towards becoming part of the broader customer experience. The question will be how payments can become an embedded part of an end-to-end business process. Today, a payment often carries limited contextual information. As standards evolve and more information, such as invoices and other business data, can travel with payments, there is an opportunity to make payments much more deeply embedded in B2B processes.

Second, I expect greater discussion around interoperability between traditional payment rails and digital assets, including tokenised deposits and stablecoins. Today, these ecosystems do not always interoperate effectively. The industry will need to address how traditional and emerging payment mechanisms can work together while maintaining appropriate controls.

Third, we may be talking about artificial general intelligence and what increasingly capable AI systems mean for financial services and payments.

The banks that succeed in this environment will be those willing to take controlled risks, move quickly, make the right technology decisions, and collaborate with partners. The world is evolving too quickly for any organisation to have all the answers. The journey matters, but choosing the right partner for that journey may matter even more.

About author

Deepak Gupta is Chief Product and Delivery Officer at Volante Technologies. A career innovator in cloud and software-as-a-service, his industry experience includes senior executive roles as General Manager at CoreLogic, President and CEO at Workstream, and CEO at iSpheres. Deepak was also SVP and GM of PeopleSoft and Chief Architect of Oracle’s SaaS business units, where he led the organisations’ transformation from on-premises enterprise software providers to SaaS leaders.

 

About Volante Technologies

Volante Technologies is the trusted payments modernisation partner for financial businesses worldwide. Volante's ISO-native, cloud-native Payments as a Service platform, powered by Agentic AI and low-code, lets institutions rapidly modernise and operate global payment rails, processing millions of mission-critical transactions and trillions in value daily. Customers include four of the top five global corporate banks, seven of the top ten US banks, and two of the world's largest card networks.