AI is reshaping payments, but banks must decide whether to repair legacy systems or rebuild infrastructure to stay competitive and compliant.
Artificial Intelligence (AI) is empowering agile startups to challenge the established banks by delivering payment solutions that are faster, cheaper and more efficient. In response, the major banking institutions are making significant investments in AI technology. However, for the most part, they are spending their money on adding AI to their existing systems, rather than redesigning those systems.
Writing in the launch edition of the ‘Journal of Financial Services‘, Alan Verschoyle-King (U.K. Payments Practice Lead for Projective Group) and Stephen Peters (Head of Enterprise Payment Solutions at FIS) argue that we have reached a pivotal moment when the major financial institutions must choose between either repairing or rebuilding their core infrastructure. The authors consider how AI will shape the future operating model of payments and examine the operational excellence necessary to make the industry’s transformation a reality.
AI may be transforming payment operations across many areas, but three in particular are in the spotlight.
- Fraud detection. AI models have improved the process somewhat by allowing banks to detect suspicious activities in real-time, but this has not been entirely effective, with online payment fraud surpassing U.S. $44 billion last year.
- New payment methods such as biometric payments, face-pay, and voice-activated payments.
- Personalising and improving the payment process. AI helps bank employees spend less time reviewing and approving payments and more time providing better, value-generating services for clients and the banks themselves, whilst reducing human intervention and eliminating errors.
AI is becoming deeply embedded in many essential areas of payment. Generative and traditional AI are no longer add-ons that automate isolated tasks. They’re becoming the decision-making fabric of routing, risk, liquidity, and client experience. Stronger authentication at the front end, real-time fraud interdiction in the middle, and self-tuning liquidity management at the backend all draw on continuous-learning systems. Banks that fail to embed this layer will face the same margin and speed squeeze that mobile-first FinTechs triggered a decade ago.
Most large institutions still process high-volume flows on monolithic stacks designed for batch processing, not real-time events. Exception queues, low straight-through processing rates, and opaque status messages remain stubbornly high despite years of “digital transformation”. Wrapping cores with middleware bought time but did not eliminate technical debt. AI-centric operations require deeper refactoring, payment operations demand observability of every event, risk and compliance systems need to consume those events in real-time, treasury and liquidity engines must predict funding needs seconds ahead, and data platforms need to broadcast ISO 20022-rich messages tagged with UETRs. Without this rewiring, AI will only operate at the edges.
Probabilistic outputs, model drift, and opaque reasoning paths unsettle regulators and boards. Supervisory frameworks from the ECB, PRA, and DORA already require auditable controls around every production model. Institutions that scale safely implement cross-functional AI councils spanning risk, compliance, operations, and technology. They maintain single inventories that auto-record every model release through CI/CD pipelines. They use time-boxed approval cycles, so governance doesn’t become an organisational parking brake. And they embed human-override kill switches into workflow tools, rather than burying them in source code.
The authors suggest these practical steps for businesses:
- Start with workable plans that will evolve
- Choose the right tool for the job
- Expect AI models to need to be refreshed
- Put in place governance that enables, rather than blocks, progress
- Invest in understanding what’s really happening
- Measure things that matter to leadership
- Modernise thoughtfully, rather than aggressively
The AI-operated payments infrastructure is not a theoretical proposition. Every component already exists in production somewhere—such as event streaming, ISO 20022 data, agent-based orchestration, and model telemetry. What differentiates leaders is treating AI as infrastructure, not experimentation.
The window for decisive action is narrowing. The banks and financial institutions that adopt AI and make fundamental changes to their infrastructure will thrive, while those treating it as just another software upgrade may struggle to keep pace with their more agile competitors.
To read the full article, download the Journal of Financial Services.
