Legacy IT debt is slowing AI adoption across the UK public sector
Outdated systems are still dragging on AI adoption and broader digital change across government. New research highlights the scale of technical debt in public sector estates and the real-world friction it creates for delivery, risk, and cost.
Cloudhouse's State of Technical Debt 2025 report surveyed 250 UK IT leaders across government, finance, and manufacturing. The core message is clear: legacy platforms are a structural barrier to modernisation, and they're holding back AI initiatives in most organisations.
Key findings at a glance
- Widespread Windows technical debt: 84% of government organisations report Windows-related debt (vs. 92% in manufacturing and 89% in finance).
- AI hindered by legacy: Three in five organisations say legacy systems are already slowing AI adoption, with most struggling to integrate AI tools with older infrastructure.
- Funds pulled into maintenance: Around 45% say budgets are being diverted from innovation to keep legacy running, delaying cloud moves and new capabilities.
- Operational and compliance risk: Nearly half have faced audit challenges tied to legacy IT. Outdated systems are linked to more downtime and higher cyber exposure.
- Intent outpacing investment: Most plan to modernise within two years, but only a minority have fully funded roadmaps.
Why this bites in government
Departments run large, interconnected estates built up over decades. These systems are kept alive to protect critical services, but the trade-off is rising operational risk, slower change, and fragile integrations.
AI makes that tension visible. Models and data pipelines are hard to connect to ageing apps and infrastructure. The result: pilots stall, costs rise, and benefits land late-or not at all.
This is also a policy and assurance issue. Compliance gaps, security exposure, and service resilience risks escalate over time. Government targets for digital and data-such as the UK's Transforming for a Digital Future roadmap-assume steady reduction of legacy risk, not status quo maintenance.
The blockers: tech, skills, funding
Three themes recur across the research: entrenched legacy, scarce modernisation skills, and budgets pulled into keeping the lights on. Together, they create a cycle-maintenance crowds out progress, progress needs skills you don't have, and the legacy footprint grows older and harder to touch.
Breaking that cycle takes deliberate prioritisation, staged delivery, and clear ownership of technical debt across the portfolio.
What to do in the next 6-12 months
- Build a single view of debt: Catalogue end-of-support systems, Windows dependencies, integration choke points, and data flow constraints that block AI use cases.
- Prioritise by service risk and value: Sequence upgrades where they reduce audit findings, security exposure, or unblock high-impact AI/data services. Protect frontline continuity with parallel run and rollback plans.
- Ring-fence modernisation budget: Cap maintenance spend where possible and reallocate to targeted debt paydown tied to measurable outcomes (e.g., decommission X systems, reduce vulnerabilities by Y%).
- Use proven patterns over big-bang change: Strangler patterns, API gateways, containerisation, and app virtualisation can decouple legacy while you modernise incrementally.
- Close the skills gap: Pair external specialists with internal teams and invest in training on cloud-native operations, data integration, and AI deployment. See AI for Government for resources on policy, governance, and adoption.
- Strengthen governance: Maintain a living technical debt register, set enterprise architecture guardrails, and tie funding approvals to lifecycle plans and exit strategies.
- Measure momentum: Track % of critical workloads modernised, time-to-integrate new AI tools, unplanned downtime, high-risk vulnerabilities closed, and AI pilots moved to production.
The long game
Modernisation isn't a one-off project. It's a rolling programme that reduces risk, shortens delivery cycles, and makes AI adoption routine rather than exceptional.
The report's message is blunt: intent without funding won't move the needle. Start small, ship incrementally, and keep paying down debt where it frees up the next AI capability your service users actually feel.
Your membership also unlocks:
