Britain’s biggest banks are spending billions of pounds a year keeping decades-old technology alive — and the bill is getting worse, not better. While artificial intelligence promises to reshape financial services, the uncomfortable truth is that most UK lenders are still running core operations on systems built before the internet existed. The gap between ambition and infrastructure has never been wider.
A striking new report from Ensono, a hybrid IT services provider, lays bare the scale of the problem. According to findings covered by TechRadar, UK banks are collectively spending approximately £5.7 billion annually just to maintain legacy systems. That figure doesn’t include the cost of innovation or digital transformation — it’s the price of simply keeping the lights on. And those lights are flickering.
The numbers are sobering. Ensono’s research indicates that around 92% of UK banks still rely on mainframe systems for critical functions including payment processing, account management, and regulatory reporting. These aren’t minor back-office utilities. They’re the backbone of daily operations, handling millions of transactions every hour. Some of these systems run on COBOL, a programming language that dates to 1959.
That last detail matters more than it might seem.
COBOL developers are aging out of the workforce. The average COBOL programmer in the UK is well past 50, and universities stopped teaching the language decades ago. Banks face a ticking clock: the people who understand how these systems actually work are retiring, and there aren’t enough replacements coming through the pipeline. It’s a knowledge drain that no amount of documentation can fully offset. When a critical mainframe fails at 2 a.m., you need someone who knows the code — not someone reading a manual.
The Ensono report, as detailed by TechRadar, found that 67% of UK banking IT leaders cite legacy modernization as their top strategic priority. But priority and progress are different things. Only 28% said they had made significant headway on replacing or upgrading core systems in the past three years. The rest are stuck in planning phases, pilot programs, or — most commonly — stuck dealing with the constant maintenance demands of their existing infrastructure.
Why can’t banks just rip out the old systems and start fresh? Because the risk is enormous. Core banking platforms are deeply entangled with every product, every channel, every compliance process. A botched migration doesn’t just cause inconvenience — it can freeze customer accounts, disrupt payments, and trigger regulatory action. TSB’s catastrophic IT migration in 2018, which locked nearly two million customers out of their accounts, remains a cautionary tale that echoes through every boardroom in the City.
So banks proceed cautiously. Too cautiously, some argue.
The arrival of generative AI has sharpened the urgency. Financial institutions want to deploy AI for fraud detection, customer service automation, credit risk modeling, and operational efficiency. But AI models are only as good as the data infrastructure feeding them. Legacy systems, with their siloed databases and batch-processing architectures, are fundamentally ill-suited for the real-time data flows that modern AI requires. You can’t bolt a Ferrari engine onto a horse cart.
According to recent reporting from Reuters, major global banks including HSBC, Barclays, and Lloyds Banking Group have all announced significant AI investment programs in 2024 and 2025. HSBC alone has committed over $4 billion to technology spending, with AI figuring prominently. But executives at these institutions have acknowledged, in varying degrees of candor, that legacy infrastructure remains the single biggest impediment to deploying AI at scale.
The competitive pressure isn’t just internal. Digital-first challengers like Monzo, Starling, and Revolut were built on modern cloud-native architectures from day one. They don’t carry the burden of 40-year-old mainframes. Their cost-to-serve per customer is a fraction of what traditional banks pay. And they can ship new features in weeks, not months. That speed advantage compounds over time.
Traditional banks know this. It haunts them.
The Financial Conduct Authority and the Prudential Regulation Authority have both signaled increasing concern about operational resilience in the UK banking sector. New rules that took effect in March 2025 require firms to demonstrate they can withstand severe operational disruptions, including technology failures. For banks running on aging mainframes with limited redundancy, meeting these standards is expensive and complex. Some industry observers believe the regulatory pressure will ultimately force modernization more effectively than any business case ever could.
There’s a financial paradox at work here that deserves attention. The £5.7 billion annual maintenance figure cited in the Ensono research represents money that produces no competitive advantage whatsoever. It’s purely defensive spending — the cost of not failing. Every pound spent patching a 30-year-old system is a pound not spent on innovation, customer experience, or market expansion. Over a decade, that’s nearly £60 billion in aggregate that UK banks will have spent just treading water.
And the costs aren’t purely financial. Legacy systems contribute to slower onboarding processes, clunkier digital experiences, and longer resolution times for customer complaints. In an era when consumers expect app-based banking to work as smoothly as ordering food delivery, these friction points erode trust and loyalty. Younger customers, who’ve grown up with Monzo and Starling, have little patience for systems that require branch visits or multi-day processing times.
Some banks are attempting a middle path. Rather than wholesale replacement, they’re pursuing what the industry calls “strangler fig” architecture — gradually wrapping new microservices around legacy cores, redirecting functions one by one until the old system can eventually be switched off. It’s a sound approach in theory. In practice, it requires years of disciplined execution and sustained investment, both of which are hard to maintain when quarterly earnings pressure and shifting executive priorities intervene.
Barclays has been relatively transparent about its modernization efforts, investing heavily in cloud migration and API-based architectures. Lloyds Banking Group has spent over £3 billion on its digital transformation program since 2018. NatWest has similarly committed to reducing its mainframe footprint. But none of these banks would claim the job is anywhere close to done.
The talent problem intersects with the technology problem in ways that amplify both. Banks need engineers who understand legacy systems well enough to migrate away from them, and simultaneously need developers skilled in cloud-native, AI-ready architectures. These are different skill sets, and the market for both is fiercely competitive. Fintech firms, Big Tech companies, and consultancies are all fishing in the same talent pool. Compensation packages at banks, while generous by historical standards, often can’t match what Google or a well-funded startup offers.
There’s also a cultural dimension. Large banks are hierarchical, process-heavy organizations. Software engineers who thrive in agile, fast-moving environments often find the pace and bureaucracy of a major bank stifling. Retention is as much a challenge as recruitment. One senior technology executive at a top-five UK bank, speaking at a recent industry conference, put it bluntly: “We’re not just competing for talent with other banks. We’re competing with every company that writes software.”
The AI angle adds another layer of complexity. Banks are under pressure from boards, investors, and regulators to demonstrate AI capability. But deploying AI responsibly in financial services requires clean, well-governed data — something legacy systems are notoriously bad at providing. Data trapped in mainframe silos often lacks standardization, is poorly documented, and may contain decades of accumulated inconsistencies. Cleaning and integrating this data is unglamorous, expensive work. But it’s a prerequisite for any meaningful AI deployment.
Recent analysis from The Financial Times has highlighted how European banks broadly lag their American counterparts in technology spending as a percentage of revenue. JPMorgan Chase spends roughly $17 billion annually on technology — more than many European banks’ entire operating budgets. That spending gap translates directly into capability gaps, particularly in AI and data analytics.
The UK government has signaled awareness of the issue. The Treasury’s fintech strategy, updated in late 2024, includes provisions aimed at encouraging legacy modernization in financial services, including potential tax incentives for cloud migration investments and a proposed regulatory sandbox for core banking system replacements. Whether these measures will move the needle remains to be seen. Policy signals are encouraging. Execution is what matters.
For now, the status quo persists. UK banks continue to pour billions into maintaining systems that were state-of-the-art when Margaret Thatcher was in Downing Street. They know these systems need replacing. They know AI adoption depends on it. They know challengers are gaining ground. And yet the sheer complexity, risk, and cost of modernization keeps most institutions moving at a pace that feels, to outside observers, almost geological.
The banks that crack this problem — that find a way to modernize without catastrophic disruption — will define the next era of British financial services. The ones that don’t will find themselves paying ever-increasing maintenance bills for systems that fewer and fewer people understand, while nimbler competitors eat their lunch.
That’s not a prediction. It’s arithmetic.