UK banks are spending approximately £3.3 billion every year simply to keep legacy core systems running – roughly a quarter of their total IT budgets. Nearly half acknowledge that these platforms struggle to keep up with everyday demand and strategic priorities.
For decades, this cost has been accepted as the price of stability. Core systems process millions of transactions daily, underpin regulatory compliance, safeguard customer trust, and keep money moving. Replacing them has long been seen as too risky, too costly, or too disruptive.
Article continues below
SVP, Solutions and Data Tribe Head at Mphasis.
AI is accelerating the pace of software change. Tools that speed up coding, automate testing, and support documentation are compressing development cycles. Business stakeholders increasingly expect faster releases, more personalized services, and quicker decision making.
Yet while AI capabilities are advancing rapidly, many banks’ application delivery models have not kept pace. Teams remain constrained by manual, fragmented practices built for a slower era. This gap between what AI makes possible and what legacy environments can realistically support is becoming impossible to ignore.
The hidden fragility of delivery models
Legacy systems are only part of the challenge. The deeper issue lies in how change is delivered around them.
In many banks, requirements are not unified or consistently defined. Business logic is embedded in disparate applications, duplicated across silos, or interpreted differently by multiple teams. Institutional knowledge often resides with a small group of long tenured experts rather than being captured in shared, structured assets.
Over time, this fragmentation creates layers of technical and operational debt that are difficult to see and even harder to manage. Introducing AI tools into this environment does not automatically resolve these issues. In fact, without clear requirements and harmonized rules, AI can amplify them.
AI generated outputs are only as reliable as the inputs they are given. When business logic is ambiguous or inconsistently defined, AI simply reproduces that inconsistency at scale. Teams may appear to move faster, but releases remain brittle, unpredictable, and high risk because there is no stable foundation beneath the surface.
The result is a familiar paradox: banks invest in AI to accelerate delivery, yet change still feels slow, expensive, and fraught with compliance risk. AI has broken delivery; it has revealed just how fragile traditional ones already were.
Why legacy costs keep rising
The £3.3bn annual maintenance bill is not simply a function of outdated technology. It is a symptom of accumulated complexity.
Over decades, banks have built dense webs of integrations, custom workarounds, and point solutions around their core platforms. Each regulatory update, product launch, or merger adds new layers.
Documentation quickly becomes outdated as logic is re-written to meet immediate needs, while deep system knowledge increasingly sits with a shrinking pool of experts nearing retirement.
In this environment, even small changes can trigger cascading impacts. Testing cycles stretch because teams fear breaking systems they no longer fully understand. Release windows are tightly controlled and surrounded by risk mitigation processes. Innovation becomes constrained by the need to preserve stability.
AI shines a light on this complexity. When leadership teams see what modern, AI assisted development can achieve in less constrained environments, the contrast with their own systems becomes stark. The opportunity cost of maintaining the status quo becomes harder to justify.
From automation to intelligence led delivery
If AI is exposing the limits of legacy systems and outdated delivery models, the answer is not simply more automation. What is required is intelligence led delivery.
Intelligence led delivery embeds business logic and decision context directly into the software development lifecycle. Instead of allowing rules drift across multiple codebases or relying on individual interpretation, organizations centralize and codify them in structured, reusable formats.
Requirements become standardized and traceable from business intent through to implementation and testing. Decision logic becomes explicit, version controlled, and auditable.
With this foundation in place, AI operates on much firmer ground. Code generation aligns with approved business rules. Testing automatically maps to defined decision paths. Impact analysis becomes more predictable because dependencies are visible and understood.
This reduces rework and risk while shortening time to value through clarity and control rather than speed alone.
Scaling AI responsibly
For financial institutions, speed cannot come at the expense of resilience, security, or regulatory compliance. Intelligence-led delivery places decision logic at the heart of the lifecycle making change more controlled, observable, and explainable.
As AI adoption increases, regulators are paying closer attention to how automated decisions are designed, implemented, and governed. Organizations that cannot clearly explain how business rules are applied across systems face heightened scrutiny.
Making logic explicit and traceable helps address these concerns while still enabling innovation.
This approach also supports talent evolution. By reducing reliance on tribal knowledge and manual interpretation, teams can focus on higher value work: refining business logic, improving customer experiences and exploring AI enabled services.
A practical path forward
Transformation does not require a wholesale replacement of core systems. Few banks can afford that level of disruption. Progress can begin with delivery models themselves.
By identifying high change areas, codifying business rules, and aligning AI tooling with structured decision management, banks can gradually reduce dependency on fragile processes. Over time, maintenance overhead decreases, release confidence improves, and a realistic path to incremental modernization emerges.
The billions spent maintaining legacy cores represents both a burden and an opportunity. Redirecting even a portion of that investment toward intelligence led practices enables a shift from defensive maintenance to proactive innovation.
AI is neither the enemy of legacy systems nor a silver bullet. It is a catalyst, revealing weaknesses built up through years of incremental change. Banks that rethink how they deliver software will be best positioned to scale AI safely and sustainably.
Those that do not may find that the greater risk lies not in adopting AI too quickly, but in holding on to delivery models that can no longer support the pace of modern banking.
We’ve featured the best AI chatbot for business.
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
