The hidden AI revolution happening inside payments

Irene Skrynova, CEO of Global Payments at Unlimit, explains how AI creates value through payment optimisation, cross-border efficiency, and the rise of agentic commerce.

AI in payments is something constantly in the news and on everybody’s minds. But the use cases vary from fraud scoring to entire decisioning systems. Where does Unlimit see AI actually making a difference today, and where does the hype go against reality?

The most useful AI in payments today is the AI you never see. The public conversation is fixated on the shopping agent – the assistant that buys things for you. But the real impact right now sits one layer down, inside the transaction itself. Payments has quietly shifted from being a processing function to being an optimisation and control function, and AI is already deep in that layer, whether or not anyone calls it 'agentic.'

Concretely, that means approval-rate optimisation, smart routing across rails, retry logic, fraud and risk scoring, and how each authorisation is presented to the network. These used to be static rule sets. They are becoming adaptive and decided per transaction in real time. Even something as mundane as merchant category classification, historically a fixed code, is becoming fluid for platforms that sell across multiple product lines, because the wrong context can quietly cost you approvals, pricing, and compliance standing. That is unglamorous work, and it is exactly where AI is paying for itself today.

Where reality diverges from the narrative is autonomy at the point of purchase. The agent that researches, decides, and pays entirely on its own, at scale, is not here yet. AI has become a primary entry point into the shopping journey, but it is not where transactions are closed. The large majority of people who use AI to research still return to a merchant's own surface to actually buy. AI is already shaping intent; it is not yet trusted to complete the purchase. So, the honest framing is this: AI as the optimiser inside payments is real and compounding now; AI as the buyer is early, and trust is what hasn't been earned yet.

In theory, AI agents making purchases on the user’s behalf, autonomously, sounds great. But this raises questions around authorisation, trust, and, of course, liability. What is your view on this? Is the infrastructure there yet, or are we just laying the groundwork for what’s to come?

We work on the assumption that our future customer is not a human. It is an agent acting on behalf of a person or a business, making decisions and settling in real time. Once you accept that, the interesting questions stop being technical and become questions of accountability. Can the agent transact? Almost. Should it, and under whose authority? That's the unsolved part.

The genuinely important work happening now is making agent-initiated transactions explicit rather than disguising them as human behaviour – registering agents, binding them to a verified identity, and carrying tokenised credentials so that every participant can see who, or what, actually initiated the transaction. The card networks are moving in exactly this direction with their agent frameworks, and that's the right instinct.

On top of that, a stack of protocols is forming, roughly along three lines: how agents discover merchants and trigger checkout, how an agent's authority to pay is established and trusted, and how value actually moves. Adoption won't be uniform; regions, merchants, and payment methods will move at different speeds. The recent friction between large platforms and autonomous shopping agents is a preview of how contested this will be, because agent access bypasses the interface where incumbents have historically controlled discovery and monetisation.

After years of investments and innovation, cross-border payments are still lacking for both businesses and consumers. Considering that Unlimit processes payments across multiple markets, what are the main challenges you experience, and how do you overcome them?

The core problem is that the world's acquiring infrastructure was designed for a human choosing a payment method, not for software optimising a decision across dozens of local rails. Cross-border looks like one problem from the outside, but it is really three stacked on top of each other: fragmented acceptance, expensive and slow settlement, and compliance that changes shape at every border.

The fragmentation is the part outsiders underestimate. In huge parts of the world, cards are not the default – alternative methods are. Each market has different systems with their own routing, settlement, and regulatory logic. Authorising a payment is the easy part; doing it efficiently, settling it predictably, and staying compliant across all of those systems simultaneously is the hard part.

The way we overcome it is to own the complexity so our merchants don't have to. That means holding dozens of local licenses, connecting natively to local rails rather than through intermediaries, and building an architecture flexible enough to route across cards, local instant-payment schemes, and increasingly stablecoins depending on what actually works best in a given corridor. The goal we're building toward is simple to state and hard to deliver: value should move across borders with the same speed and predictability as information does, and the merchant shouldn't have to think about the risks.

Expanding to a new country comes with months of preparatory work – integration, regulatory approvals, local partnerships – all before a single payment is processed. How does a company like Unlimit help businesses see through the clutter, reduce complexity, and what does ‘intelligent infrastructure’ actually mean?

Entering a market is mostly invisible work that happens before any revenue: licencing, regulatory approvals, local banking and rail partnerships, integration. For most businesses, that is months of effort per country, repeated country by country, and it is the single biggest reason ambitious companies stay regional longer than they should. Our role is to absorb that work at scale, so merchants can expand through a single relationship rather than rebuilding the same foundation in every jurisdiction.

That's the practical meaning of 'intelligent infrastructure,' and I'd separate it from the buzzword version. It isn't AI sprinkled on top. It's an architecture where the underlying rail becomes an implementation detail, where the system adapts to the market and the regulatory environment rather than forcing the merchant to. The intelligence shows up in the decisions the platform makes on every transaction: which rail to route over, how to retry, how to present the payment to the network, what context to assign it. Increasingly, these decisions are made dynamically, per transaction, which is a far more agent-like way of operating than a fixed rulebook.

So, when we talk about reducing complexity, we mean two things at once. One relationship instead of dozens, with the local depth – licenses, methods, compliance – already handled. And optimisation built into the infrastructure itself, so the merchant isn't just connected to a market but is competing efficiently in it from day one. The clutter a business should never have to see is precisely the clutter we exist to manage.

Everybody talks about the immense benefits AI will introduce to the payments world. Can you pinpoint something being built or developed where AI placed on top of cross-border payments produced an outcome that wouldn’t have been possible otherwise?

Yes – and the clearest example is a real-time, context-aware decisioning at the level of the individual transaction. In a cross-border flow, the 'right' answer changes constantly: the optimal rail, the optimal retry, the right way to classify and present a payment all depend on the corridor, the method, the merchant's mix, and the moment. A static rules engine simply cannot hold that many variables across that many markets. Treating each transaction as a decision to be optimised lifts approval rates and lowers costs in a way that handwritten rules never could at this scale. That is a genuinely new capability, not an incremental one.

The second outcome is even more telling: AI on top of programmable cross-border settlement is making entirely new business models viable. A new class of AI-native businesses is emerging, in which the commerce surface is an agent or an API, not a website. These models often depend on very high-frequency, small-value, instant transactions; and they break on infrastructure built for batch processing, manual review, and fixed fees. If every payment carries delay or conversion friction, those businesses simply don't exist. Combine intelligent routing with programmable settlement, including stablecoins with a clean fiat off-ramp across many markets, and suddenly machine-speed micro-commerce across borders becomes economically possible. That's the kind of thing that the old infrastructure couldn't have supported at all.

Last but not least, looking at the payments industry, what would be the one thing you could change about how it’s being approached – be it interoperability, regulation, standards, or even mindset?

Mindset, specifically the belief that whoever owns the interface owns the customer. The industry spent 20 years optimising the checkout surface because that's where conversion and behavioural data lived. That assumption is now quietly wrong. As decisions move upstream into agents, the interface matters less, and execution and trust matter more. A lot of strategic energy is still being spent defending a layer that is losing its leverage, and I'd redirect all of it toward being excellent at the things that are becoming decisive: routing, settlement, optionality, and verifiable trust.

The reflex right now is to try to crown a single winning protocol for agentic commerce and force everything through it. However, different regions, methods, and regulators will move at different speeds, and premature consolidation around one standard locks the market into choices it doesn't yet understand. Interoperability is the more honest goal: build to be compatible with every meaningful path, and let real-world usage decide what scales, instead of trying to predict it from a conference stage.

Put those together, and the change I'd advocate is a shift from control to enablement. The best thing the payments industry can do over the next decade is stop trying to own the customer's attention and start being the most reliable and interoperable layer beneath whatever or whoever is making the decision. Money should move as freely as information; our job is to make that true without anyone having to think about how.

Author the author

Irene Skrynova, CEO, Global Payments at Unlimit, is an international fintech powerhouse. As an award-winning thought leader, Irene is recognised for her strategic vision and transformative leadership on an international scale. She has played a pivotal role in shaping global customer success strategies and business transformation. Beyond her executive role, Irene mentors emerging talent, speaks and serves on global juries, and leads industry committees to drive inclusive financial leadership.

About the company