Pay360 2025: AI Moves from Feature to Infrastructure

A year ago I wrote about Pay360 2024 and noted that the payments industry's AI conversation was more mature than equivalent conversations in commerce and marketing. Grounded in production deployments in fraud, compliance, and operational efficiency, rather than speculative visions of consumer experience. That observation holds for 2025, but the scope has expanded considerably.

Pay360 2025, at ExCeL London on 25–26 March, landed in a specific moment: one month before Mastercard would announce Agent Pay, and roughly six months before Visa's Intelligent Commerce and the broader wave of agentic payment infrastructure would arrive. The payments industry was building toward something that was not yet fully visible in public announcements but was clearly forming in the thinking of the people in that room.

The Survey Data

The Payments Association's pre-event survey gathered insights from 6,578 payments professionals across 3,303 companies and 93 countries. AI rose from 25% to 29% as the dominant identified opportunity in payments between 2024 and 2025, with professionals across 76 of those countries naming it their top priority. That's a consistent upward trend, but what was more interesting to me was the shift in how the opportunity is being characterised. Less "AI will improve customer experience" and more "AI will change the infrastructure of how transactions work."

Worth noting, too: open banking as a named opportunity actually declined in the same survey, from 14% to 10%. That's not a retreat. It's what happens when infrastructure matures. You stop talking about the foundation and start talking about what you're building on it.

That's a significant pivot in framing. The customer experience conversation is about the layer on top. The infrastructure conversation is about what replaces the underlying plumbing.

What "Infrastructure" Means in This Context

Pay360's Day 2 agenda ran under the heading "Tech Horizons, AI, and Open Banking" — a pairing that reflects where the real interest lies. The combination of open banking APIs (which give AI agents programmatic access to bank accounts and payment initiation) and AI (which can reason about when and how to use those APIs on behalf of consumers) is the foundation of what was being discussed.

The specific concept generating the most serious technical discussion, at least in the conversations I was part of: tokenised, credentialed AI agents with pre-authorised payment access and spending rules. The idea that an AI agent can be given a budget, a set of purchase criteria, and pre-authorised payment credentials, then make purchases autonomously within those parameters without requiring the consumer to re-authenticate each transaction. That's what makes agentic commerce technically feasible rather than just conceptually interesting.

The fraud and security implications of this architecture were getting serious attention. An agent with pre-authorised payment access is also a significant target for compromise. The tokenisation approach that the major payment networks were developing (replacing actual payment credentials with single-use tokens tied to specific transaction parameters) is the security architecture that makes agent-initiated payments viable. This is not a new problem. The industry has been solving for credential theft since the first card-not-present transaction. It is an old problem with a new surface area.

Open Banking in 2025

The UK's open banking story in early 2025 was one of gradually expanding use cases building on solid API infrastructure. The direction was visible in the numbers: by November 2025, Open Banking Limited would report nearly 33 million open banking payments in a single month, against a base of more than 16.5 million live user connections across the UK. Those figures weren't public during Pay360, but the underlying trajectory was clear to anyone looking at the growth curves from 2023 and 2024.

Government-adjacent use cases had been demonstrating that open banking payment initiation works reliably at scale when the value proposition is clear and user friction is low. Commercial ecommerce remained more complicated. Open banking checkout as an alternative to card payment is technically better in several dimensions: no card network fees, real-time settlement, no risk of card data breach. It has struggled to achieve consumer adoption because "pay via bank" requires more conscious engagement than tapping a phone. The friction difference is small but real, and in checkout optimisation, small frictions have large effects.

What's interesting for the agentic commerce context is that open banking and agentic purchasing might solve each other's problems. An AI agent pre-authorised to make purchases on behalf of a consumer doesn't face the friction of a manual open banking checkout flow. It's already authenticated, already has the payment mandate, and executes without requiring conscious engagement. The friction that prevents consumer adoption of open banking checkout disappears when the consumer isn't making the individual purchase decision.

This is the kind of structural symmetry that tends to generate genuine momentum in the payments industry. Two separate problems with the same solution, arriving at the same time.

The Emerging Picture

Pay360 2025 was a conference where the participants knew they were in the final months before agentic payments infrastructure went live at scale. The Mastercard announcement a month later, the Visa and PayPal announcements later in the year, the Stripe and Google infrastructure that would follow — these were in progress during the conference, known to the people building them if not yet public.

The mood wasn't euphoric. It was something more like focused and slightly apprehensive. The financial services industry has a long institutional memory of transformative technologies that arrived with promises and left complicated legacies. The conversation at Pay360 had the texture of people who wanted to get this right, not just fast.

The security questions are real. The regulatory questions are real. What happens when an AI agent makes a payment the consumer would have refused? What happens when an agent is compromised and makes a payment that constitutes a fraud? These are not hypothetical edge cases. They are architectural questions that require architectural answers, and the payment networks were working on them.

That seems like the appropriate posture for an industry where "move fast and break things" is genuinely not acceptable. The payments infrastructure that will underpin agentic commerce is being built by people who understand the stakes. On the evidence of Pay360 2025, they were taking them seriously.

Focused and apprehensive. The right combination.