AI industrialisation in banks moves beyond pilots, KPMG data shows

AI industrialisation in banks is moving from discrete trials into the operational core of financial institutions, according to Tamsin Crossland, principal AI architect at Icon Solutions. Crossland cited KPMG research showing active use of AI in financial services rising from 30% in 2024 to 75% in 2026, a shift with direct implications for payments, fraud prevention, compliance, customer service and software development.

In an external opinion published by Finextra, Crossland said large banks are increasingly treating AI as a core capability rather than a separate innovation programme. The change is pushing institutions to reconsider their technology architecture, data access and risk controls as they try to deploy AI across multiple business functions without adding operational complexity.

What is AI industrialisation in banking?

AI industrialisation in banking refers to the use of artificial intelligence at enterprise scale, with models, data systems, governance and workflow tools built into day-to-day operations. In Crossland’s view, the emphasis is shifting from isolated applications to platforms that can support repeatable, supervised processes across the bank.

Crossland said the most consequential adoption is taking place inside banking operations rather than only in customer-facing chatbots. Banks are applying AI to payments, compliance, fraud teams and engineering work to improve efficiency, productivity and resilience, according to her assessment.

She also pointed to the emergence of agentic AI. These systems can handle multi-step workflows, gather and combine information, interact with enterprise systems and refer decisions to staff where needed. Crossland described the model as human-supervised automation, with employees retaining responsibility for approvals, oversight and regulatory judgement.

Why bank AI architecture is changing

Crossland said banks are moving away from reliance on one model or one vendor. Instead, they are assembling modular systems that combine different models, orchestration tools, governance controls and enterprise data foundations.

That structure matters because AI systems in regulated financial services need access to accurate, governed and context-specific data. Crossland identified retrieval-augmented generation, vector databases and semantic search as components that help banks connect AI tools to internal policies, procedures and operational records.

Retrieval-augmented generation, often shortened to RAG, is a method that lets an AI system retrieve relevant internal material before producing an answer. In a bank, that can reduce reliance on generic model outputs by tying responses to approved documents and data.

Crossland argued that model choice alone is unlikely to determine winners in banking AI. She said integration, orchestration and redesign of operational processes are central to whether banks can use the technology safely and at scale.

What is holding banks back?

Crossland said progress remains uneven because many banks still rely on fragmented legacy systems, inconsistent data models and limited real-time access to information. AI systems perform better when data is clean, connected, governed and rich in context, she said.

Older core banking and payments systems also pose a constraint. Crossland said integrating AI into live, mission-critical environments can be complex, costly and slow because those systems were not built around AI-supported workflows.

Regulators are also sharpening their focus. Crossland cited a joint statement from the Financial Conduct Authority, Bank of England and HM Treasury on frontier AI models and cyber resilience, which flagged new forms of AI-related cyber and operational risk as advanced models become more capable and widely used.

Crossland said banks deploying AI across the enterprise will need explainability, auditability, observability and human oversight built into platform design. Over the next three to five years, she expects AI to become more embedded in payments, compliance, fraud operations and customer service, with systems supporting real-time interpretation, exception handling and decision preparation under staff supervision.

This story draws on original reporting from Finextra Research.