The Next Phase of ISO 20022 Compliance: Preparing Payments for AI

AI is dominating the payments conversation. 45% of surveyed financial professionals identified the increased use of AI for transaction automation and fraud detection as one of the most influential trends shaping the future of payments. Its use cases span from improving operational efficiency to providing greater customer convenience. Yet, this technology depends on data quality.  

AI needs rich payment context, party information, remittance details, transaction data, and compliance-relevant fields. ISO 20022 provides a standardized structure for richer payment context, but banks must still ensure that the underlying data is complete, accurate, and preserved throughout the payment lifecycle. A recent Faster Payments Council report reinforces this point, indicating that this standard can render information fit for automation.  

The roadblock is that many financial institutions implemented stopgap measures to quickly meet ISO 20022 migration deadlines. These initiatives did not create a viable foundation for modern intelligence systems. If banks want to move from barebone compliance to ISO 20022-powered intelligence, they will need to fix the underlying structure. 

Legacy Architectures Hold AI Back 

When CHIPS completed its transition to the ISO 20022 standard in 2024, it marked a major U.S. milestone within the broader financial messaging infrastructure challenge and exposed the architectural decisions banks would need to make next. Financial institutions added translation layers, message converters, or stopgap middleware to address compliance requirements without incurring high application-rewrite costs. The ensuing Fedwire Funds Service migration and the end of the SWIFT cross-border MT/MX coexistence period prompted many similar implementation decisions. 

Now, financial institutions are operating more moving parts than their infrastructure can effectively handle. The result is that their data is flattened into legacy formats, translated away, or trapped in siloed systems. These limitations affect AI’s ability to operationalize this information safely.  

For AI to draw on this context reliably, banks will need to modernize the data and workflow architecture surrounding ISO 20022 so that richer information is preserved and operationalized rather than translated into legacy constructs. That modernization can occur progressively, without requiring every underlying system to be replaced at once. 

The Architecture AI Depends On 

To create a reliable foundation for AI, financial institutions will need messaging architectures that support multiple rails, messaging standards, and regulatory conditions, without demanding extensive custom builds.  

The distinction between ISO-compliant and ISO-native architecture is critical. In an ISO-native environment, the structured data model is embedded in the architecture and workflows rather than translated only at the network boundary. Rich payment information can therefore remain available for validation, enrichment, routing, reconciliation and exception handling throughout the payment lifecycle. Translation-based approaches may satisfy message-format requirements, but they can reduce or obscure the context on which intelligent automation depends. 

To satisfy these requirements, many banks are shifting toward API-first and cloud-native platforms, which combine to empower banks to modernize connectivity, data management and processing in stages while maintaining a consistent payment object and end-to-end trace across the transaction lifecycle. 

Taken together, these shifts allow banks to decouple financial messaging from processing, enabling each layer to modernize independently. This means payment operations teams can shift away from maintaining connectivity components and start building the secure, unified foundation that will ultimately support agentic AI. 

Putting AI to Work in Payments 

If financial institutions rebuild their ISO 20022 systems into a unified environment that AI can draw on, they can introduce semi-autonomous and autonomous actions into targeted payment workflows, beginning with assistance and expanding autonomy as performance is validated. 

The result is AI agents interacting with all parts of a transaction flow. They can analyze data as it enters a banking environment, identify payments that will trigger transaction failures, and repair them where possible. Agents can also analyze historical payment activity to determine whether a transaction is likely to trigger a failure down the line. Coupling these practices with ongoing AI monitoring can help detect any transaction anomalies and keep the necessary parties informed.  

Banks must design governance into these capabilities from the outset, even while they are still validating operational value. 

Making AI Safe for Payments 

With 91% of financial professionals concerned about AI risks, financial institutions are often hesitant to hand over complete control of live money to autonomous platforms. Establishing guardrails is therefore essential.  

Payment operations teams should define AI boundaries before banks deploy the agents into transaction workflows. They need to decide in which situations an agent can act on its own, the currency amounts it can’t exceed, and when a human must approve an action. 

Another way to improve trust is to test agents before giving them autonomy. Payment operators should run new AI agents in shadow mode, allowing them to observe live data and record how they would have interacted with it – without actually interfering. This approach allows payment operations teams to verify that an agent can successfully interact with mission-critical information before banks deploy it into production.  

If AI produces an error, human operators can correct the output, and AI can store that information to use for future transactions. This feedback allows AI to continue learning and improving. If performance dips, banks should also have safeguards in place that can deactivate an agent, preventing it from processing additional payments. These are the kinds of controls that give financial institutions greater confidence in AI suggestions and allow them to trust the technology to handle transactions safely. 

ISO 20022 Compliance Was the First Step, Not the Last 

ISO 20022 compliance has drawn a divide. There are those who solved just for compliance, and there are those who used the deadlines as an opportunity to upgrade their underlying structure. Banks that modernized the underlying data architecture are significantly better positioned to support production-grade payments of AI than those still relying primarily on translation and temporary compatibility layers. 

Those who upgraded their infrastructure, rather than applying temporary layers and converters, have taken the first step toward modernization. They have built the foundation for the future of agentic payments operations and real-time action. This investment will only become more valuable as new advancements rise throughout the industry.