Why Government is failing to scale AI across its departments

Exciting technologies such as AI (agentic, generative, and machine learning) have huge potential to deliver value to the UK Government. But AI will not transform government unless we fix the foundations. That starts with funding the unglamorous work of retiring legacy systems, the key to unlocking everything else.

In my experience, Government Digital, Data and Technology teams don’t lack ideas or exemplars, they already know where AI can add value. What holds them back is the culture and operational processes built around legacy systems: outdated workflows, data locked in proprietary platforms, a non-digital mindset, hierarchical funding decisions, and slow procurement that makes it hard to experiment or move at pace.

In many ways, the technology itself is the easiest part to fix. Modern systems can be configured, iterated, and improved continuously. They do not require negotiation, persuasion, or cultural buy-in, they simply need clear problems they are designed to solve.

The more complex challenge is people. Over time, workforces have built entire processes around legacy systems. Changing those systems requires people to think differently, adopt new ways of working, collaborate across boundaries, and trust unfamiliar tools. This creates resistance: people become risk-averse, disengaged, or sceptical, often attributing challenges to the technology rather than the underlying change.

Introducing AI intensifies this challenge. It represents not just a new tool, but a fundamentally different way of working – one that can feel unfamiliar and even threatening. People may question its feasibility, resist new uses of data, or doubt its relevance. This human dimension is the true legacy barrier, and addressing it requires sustained investment in skills, support, and change leadership across digital, data, HR and transformation functions.

At the same time, structural issues persist. Government funding for innovation is often ring-fenced, leaving core legacy systems untouched. This leads to siloed pilots and proofs of concept that cannot integrate with existing systems and therefore cannot scale. Similarly, long strategy cycles and slow approvals mean initiatives risk becoming outdated before they are delivered, while funding decisions are not always made by those closest to operational need.

Ways AI is transforming government for both staff and citizens

Customer service, from chat tools that guide citizens through complex processes to “agentic AI” that completes tasks on their behalf, reduces friction for citizens while freeing up staff to focus on more complex or sensitive cases.

In regulatory and compliance functions, AI can and is in small pockets, analysing large volumes of data to identify anomalies, flag risks, and prioritise interventions. Rather than manually reviewing cases, teams could focus on higher-value decision-making, supported by AI-generated insights.

Within policy and strategy, AI could accelerate evidence gathering and analysis, synthesising consultation responses, modelling potential impacts of policy options, and generating scenario-based roadmaps. This would enable faster, more informed decision-making while maintaining transparency.

AI also has significant potential to improve internal productivity. Tools could support drafting briefings, summarising documents, automating routine correspondence, and surfacing relevant information from across departments. Over time, this could reduce administrative burden and allow civil servants to spend more time on delivery and outcomes.

Beyond individual use cases, AI could generate automatic roadmaps for digital transformation, identifying inefficiencies across services, mapping dependencies between systems, and suggesting prioritised interventions. This creates a more proactive, system-wide approach to reform rather than isolated improvements.

But this will only happen if leaders, not just technical specialists, are upskilled. “AI for all” should mean every civil servant has the tools and confidence to use AI in their role, alongside clear guardrails on responsible use. Without this, the benefits will remain uneven and limited to pockets of innovation rather than embedded across the system.

Data, governance, and risk

Progress is being made on data ownership and privacy-enhancing technologies, but governance is still patchy, and risk appetite remains low. Too many processes are designed for caution, not innovation. Unlocking datasets and modernising governance must go hand-in-hand with replacing legacy technology.

Scaling AI the right way

The danger is that AI is used to digitise outdated processes rather than redesign services from the ground up. Ministers and leaders should adopt a test-and-learn approach: experiment, build exemplars, and scale only what delivers real value.

We must invest across all departments so we can deliver all these potential transformations across all of Goverenment.

To remove these barriers, we need:

• Targeted funding to modernise legacy systems and unlock data

• Procurement processes aligned with agile, flexible delivery

• Investment decisions empowered at the right level, informed by real user and operational needs

• Sustained support for workforce transformation, including skills, change management, and cross-functional collaboration

The bottom line is that if we can’t do all of these, then we will fail to realise the benefits of our data and AI will remain confined to pilots and proofs of concept. The real prize is not experimentation, but scale. To achieve that, government must move beyond isolated innovation and tackle legacy in all its forms, enabling AI to deliver meaningful, system-wide impact.

Esther O’Sullivan will be speaking at next months Think AI for Government conference along with other AI and digital leaders from across government. You can register to attend here.