Banks face a curious contradiction: artificial intelligence could revolutionize how they operate, yet they're holding back from deploying it where it matters most. While AI excels at automating routine tasks like customer service and document processing, financial institutions remain skeptical about using AI for critical decisions such as mortgage underwriting and trading, according to analysis from the Bank of England. The tension stems from a fundamental mismatch between how AI works and what banking demands.

Why Are Banks Hesitant About AI in High-Stakes Decisions?

The core issue is interpretability and predictability. Banking is not advertising. If an AI algorithm decides which advertisement you see, the reasoning behind that choice matters little. But if an AI system denies your mortgage application, freezes your bank account, or flags you as a potential money launderer, someone needs to explain why. That requirement sits uncomfortably beside probabilistic AI systems, which make decisions based on probability patterns rather than deterministic rules. Their internal reasoning can be difficult to interpret, and their outputs are not always predictable.

The Bank of England's Financial Policy Committee found that advanced AI had yet to be widely adopted for core decisions such as underwriting and trading. Financial firms judged that problems around interpretability and predictability could outweigh the gains in these high-risk areas. In other words, banks want the intelligence but are less enthusiastic about the unpredictability that comes with it.

What Are the Real Strengths of AI in Banking?

Banks occupy one of the most privileged positions in the economy: they sit between identity, money, data, and trust. They know what customers earn, what they spend, where they spend it, when they spend it, what they borrow, what they save, and what they invest. For decades, banks have struggled to turn that mountain of data into useful intelligence. AI changes the equation because it can identify patterns across billions of interactions that humans and conventional systems would struggle to find.

The immediate AI opportunity is efficiency. The Bank of England reports that firms are already reporting productivity improvements from AI, particularly in software development, finance, administration, and customer service. Importantly, the largest gains occur when skilled humans validate and refine AI outputs. The first generation of AI banking will not be AI replacing people; it will be people using AI replacing people who don't.

Beyond efficiency, AI enables a profound shift in how banks relate to customers. Imagine a bank that understands your cash flow, bills, savings, mortgage, investments, pension, tax obligations, and spending patterns continuously. Instead of waiting for you to discover that you have a financial problem, it sees the problem developing and acts before it arrives. Your bank stops being somewhere you go to manage money and becomes something managing money around you.

What Weaknesses Prevent Banks From Fully Adopting AI?

Putting AI into a bank is not the same thing as putting AI into a technology company because banks carry decades of technology history. Core banking platforms, payments engines, risk systems, compliance databases, customer records, and product platforms have often been built at different times, using different architectures, with data scattered across hundreds or thousands of systems. AI may be intelligent, but it cannot magically repair bad architecture.

Banks may have more financial data than most industries, yet they find it among the hardest to use coherently. The principle of "garbage in, garbage out" has not disappeared because we invented large language models. If anything, AI magnifies the problem. Give a human poor data and you may get one poor decision. Give an autonomous system poor data and you can generate millions of poor decisions at machine speed.

How Can Banks Overcome These Barriers?

• Data Architecture: Banks must modernize fragmented legacy systems and consolidate data across hundreds or thousands of platforms to create coherent datasets that AI can reliably analyze without introducing errors at scale.
• Explainability Standards: Develop AI systems that can clearly explain their reasoning for high-stakes decisions like lending and fraud detection, moving beyond probabilistic black boxes toward interpretable models that regulators and customers can trust.
• Human Validation Layers: Implement workflows where skilled humans review and refine AI outputs before deployment, ensuring that the first generation of AI banking augments human judgment rather than replacing it entirely.

What Does the Future of AI Banking Look Like?

The biggest opportunity from AI is not reducing the cost of banking; it is reinventing banking itself. Banks are moving from generative AI, which creates content, toward agentic AI: systems that do not merely answer questions but can plan and execute sequences of actions. The Bank of England describes this as an inflection point. An intelligent financial agent can watch, analyze, predict, recommend, and increasingly act. Agentic AI can move money, negotiate transactions, manage liquidity, rebalance portfolios, detect fraud, originate credit, and interact with other agents without waiting for a human being to press a button.

"The financial system could therefore evolve towards one operating more autonomously, with agents transacting for consumers and merchants and potentially devising and executing strategies in financial markets," noted Sarah Breeden in Bank of England analysis.

Sarah Breeden, Bank of England

Today, if you ask your banking app whether you can afford a holiday, it shows you your balance. A smarter bank analyzes your salary, mortgage, bills, and savings and tells you whether you can afford the holiday. An agentic bank could go much further. It could identify how much you can afford, move surplus cash into savings beforehand, search for appropriate travel options, optimize foreign exchange, arrange insurance, schedule payments, and automatically adjust your monthly finances.

For small and medium-sized enterprises, an AI agent could manage invoices, working capital, foreign exchange exposure, cash forecasting, credit lines, supplier payments, tax reserves, and investment of surplus liquidity continuously. Banking becomes embedded into the operating system of the business.

The opportunity becomes even larger when AI meets programmable money, tokenized assets, stablecoins, and digital currencies. An AI agent that can reason and move money is transformative. The Bank of England is already examining agentic payments because autonomous systems could initiate and optimize financial flows at far greater speed and scale. However, there is tension at the heart of this development: AI is probabilistic, while payment infrastructure demands deterministic, legally certain outcomes. Solve that problem and banking enters an entirely different financial world.

Machine-to-machine commerce becomes possible. Software buys services from software. Cars pay charging stations. Supply chains negotiate payments automatically. Corporate treasury agents move liquidity around the world continuously. Money becomes increasingly programmable and autonomous. That may be one of the largest opportunities banking has seen since the invention of electronic payments.