New technologies are usually discussed in terms of what they can do: their intelligence, their speed and their ability to remove friction. Far less attention goes to what must sit underneath them for them to work safely, at scale and in the real world. According to Tieto, that blind spot is now emerging in payments as agentic AI gains ground.
Over the past year, AI agents have moved quickly into invoice matching, payment exception handling and even autonomous transaction initiation. Forrester has identified agentic AI as a key differentiator in accounts payable automation, with leading teams achieving 60–80% touchless invoice processing.
Tieto, which works at the meeting point of corporates, banks, ERP systems and payment providers, argues that speed alone is not the full picture. If underlying data is inconsistent, bank formats vary or ERP connectivity is patchy and manually maintained, AI simply reaches a faster decision on a weak foundation. When money is moving, a faster wrong decision is no improvement.
For years, corporates and ERP providers have relied on point-to-point connections to each bank and payment provider, Tieto said. That approach worked when counterparties were few and change was slow. Today, the ISO 20022 migration, demand for real-time visibility, a widening range of payment methods and the rise of data-hungry AI systems are combining to create complexity that traditional connectivity was never built to handle.
Connectivity asks whether one system can talk to another. Orchestration asks whether an expanding network of banks, ERPs, corporates, payment providers and AI agents can interact in a governed, auditable and reliable way, without each participant rebuilding the same plumbing. It allows firms to connect once and reach many, rather than re-engineering integrations whenever a bank changes a format, regulation shifts or a new payment rail appears.
Tieto highlights three reasons this matters more as AI advances. First, auditability: when an AI agent approves a payment, the trail of intent must be at least as clear as with a human approver. Second, standardisation: AI can learn one bank’s quirky format, but learning hundreds is inefficient, which is why standards such as ISO 20022 are essential for scaling automation. Third, exceptions: AI handles the routine majority well, but the messy edge cases are where infrastructure quality decides whether AI resolves, escalates or quietly gets it wrong.
The company believes the Nordics are well placed to lead, thanks to high digital maturity, established payment infrastructure and broad adoption of e-invoicing and open banking. That gives the region a chance to build governance in from the start, rather than retrofitting trust after a major incident forces the issue.
For more, read the full story here.
Copyright © 2026 FinTech Global
