AI in payments is no exception
Over the past year, I've watched the agentic AI narrative accelerate rapidly when AI agents handling invoice matching, resolving payment exceptions, even initiating transactions autonomously. Forrester recently noted that agentic AI is becoming a key differentiator in accounts payable automation, with best-performing teams reaching 60–80% touchless invoice processing. That's a meaningful signal of where the market is heading.
But in my day-to-day work on financial connectivity sitting at the intersection of corporates, banks, ERP systems, and payment providers I keep coming back to a question that gets far less airtime: what is AI actually standing on when it makes these decisions faster?
Part of the story that gets skipped
An AI agent that resolves payment in seconds instead of hours is undeniably valuable. But that speed is only as good as what's feeding it.
If the underlying data is inconsistent, if formats differ from one bank to the next, if connectivity to ERP systems is patchy or manually maintained, then AI isn't solving the problem it's just making a faster decision on top of a shaky foundation. And a faster wrong decision is obviously not better than a slower one, especially when money is moving.
This is the part of the agentic AI conversation that I think deserves more attention:
- "AI doesn't remove the need for strong financial infrastructure. It raises the stakes on it. "
The smarter the decision-making layer becomes, the more it depends on clean, standardized, well-governed data moving reliably underneath it. Intelligence without infrastructure is just guesswork with better marketing.
From connectivity to orchestration
For years, the standard approach for corporates and ERP providers has been to build individual connections point-to-point "pipes” to each bank and payment provider they work with. It was a workable model when the number of counterparties was small, and the pace of change was slow.
That model is breaking under its own weight.
When I look at what's happening across payments and financial connectivity, several shifts are happening at the same time: the ISO 20022 migration, growing expectations for real-time visibility, an expanding landscape of payment methods, and now AI systems that depend on trusted, real-time financial data.
The result is a level of operational complexity that traditional connectivity alone was never designed to solve. And that, to me, signals a broader shift:
-"We are moving from basic financial connectivity to true financial orchestration."
The difference matters. Connectivity solves the question "can System A talk to System B?" Orchestration solves a bigger question: "can a growing, shifting network of banks, ERPs, corporates, and payment providers and increasingly AI agents acting on their behalf interact with each other in a way that is governed, auditable, and reliable. Without everyone having to build and maintain their own version of the same plumbing?"
Orchestration is what allows a corporate or software provider to connect once and reach many rather than rebuilding a new integration every time a bank changes a format, a regulation shift, or a new payment rail emerges.
Why this becomes more - not less important with AI
It would be easy to assume that as AI gets more capable, infrastructure becomes less important so that intelligence can simply compensate for messy underlying systems. I think the opposite is true, for three reasons:
1. Auditability becomes non-negotiable. When a human approves a payment, there's a clear, traceable moment of intent. When an AI agent initiates or approves a transaction, that trail needs to be reconstructed differently and regulators, auditors, and banks will expect it to be every bit clear, if not clearer. Orchestration platforms that provide structured, auditable data flows aren't a nice-to-have in this world; they're the baseline requirement for AI to be trusted with financial decisions at all.
2. Standardization is what makes automation scale. An AI model can learn to interpret one bank's quirky file format. It's far harder and far less efficient to ask it to learn hundreds of different formats across hundreds of banking relationships. Standardization (ISO20022 being a prime example) isn't just a compliance exercise; it's what allows automation, AI-driven or otherwise, to actually scale across a network rather than being re-engineered for every new connection.
3. Exceptions are where trust is won or lost. AI is genuinely good at handling the "easy" 80% of transactions and exceptions. It's the remaining, messier cases that don't fit a clean pattern where infrastructure quality determines whether AI can even attempt a resolution, or whether it has to escalate, stall, or worse, get it wrong quietly. The quality of the orchestration layer directly determines how much of that 80% can genuinely become touchless, and how safely the rest gets handled.
A Nordic advantage, if we choose to use It
The Nordics are unusually well positioned to lead on this. We have high digital maturity, mature and well-adopted digital payment infrastructure, and widespread use of e-invoicing and open banking interfaces. Few regions have this combination of readiness.
That gives us a genuine opportunity not just to adopt AI in payments, but to help define how it's done responsibly: with the right standards, the right governance, and the right orchestration layer underneath it, rather than retrofitting trust and oversight after the fact, which is historically how we tend to handle new technology.
The question worth sitting with
The question is not whether AI will transform payments and financial processes it will, and in important ways, it already has.
The more interesting question is this: as AI agents take on more of the decision-making in payments and accounts payable matching invoices, resolving exceptions, flagging differences where are the biggest gaps in making that safe to scale? Is it data quality? Format standardization? Legacy infrastructure that was never designed for this kind of throughput? Or something else entirely that hasn't surfaced yet?
I don't think there's a single clean answer. But I think it's a conversation worth having now, among the people building the infrastructure, the banks relying on it, and the regulators who will eventually need to make sense of it rather than after the first major incident forces the issue.
If you work in treasury, ERP architecture, banking, or payments and have a view on this, I'd genuinely like to hear it. Feel free to get in touch. We in Tieto supports Nordic companies with the payment connectivity that helps financial processes run reliably as they scale.
