The National Audit Office (NAO) plays an important role in ensuring that government departments are delivering value for money for taxpayers. Over the years, it’s tackled some of the thorniest IT-related issues, including digital transformation in government and cyber security and resilience.
So it should come as no surprise that it has also turned its attention to how artificial intelligence (AI) might be used to improve public services. A 2024 report – Use of Artificial Intelligence in Government – highlighted the potential benefits of this game-changing technology.
But it also raised a number of “risks and concerns”, including the need to address fundamental barriers to AI adoption, such as legacy systems and data access and sharing.
From AI ambition to operational reality
Of course, the report was written at a time when the public sector was only really just beginning its AI journey. Two years on, and at the beginning of 2026, the head of the NAO spoke to MPs and civil servants about the need to get better results and outcomes for taxpayers.
Gareth Davies revealed, for instance, that the NAO is looking to provide an update on the use of AI and whether the expected benefits in efficiency and service quality are being realised
“Departments are identifying some promising results,” he said. “In our report on police productivity, we highlighted how the use of AI-enabled tools for redacting sensitive information from videos, CCTV and audio recordings had saved significant time for police officers – with the potential to save about 11,000 days a month if adopted nationally.”
But he also sounded a note of caution, saying, “Of course, we are also seeing plenty of evidence that AI comes with risks to accuracy which must be managed if public trust is to be secured. We will take a pro-innovation approach to our work, recognising that not all investments will pay off, and helping build the evidence base for high-value interventions.”
Why public sector AI pilots continue to stall
He makes an interesting point. According to the Project Management Institute (PMI), between 70-80% of AI proof-of-concept (POC) projects fail to translate into successful real-world applications, thanks in part to vague or unrealistic expectations or misaligned goals.
But the PMI also noted that the quality and supply of good data were also to blame for many project failures.
Indeed, many public sector organisations are now discovering that getting real value from AI is not simply about adopting the technology. Instead, it depends greatly on whether their existing systems, processes and infrastructure are capable of supporting it effectively.
This chimes with my own experience. Talking to people across different government departments and agencies, I know that managing growing volumes of data across a mix of legacy systems, cloud platforms and hybrid environments is becoming increasingly challenging.
Why visibility and observability are becoming critical to AI success
Of course, operating in complex environments has always been a challenge. But the growing use of AI is significantly increasing the pressure on already stretched IT and operational teams.
After all, AI systems rely heavily on access to accurate, timely and well-managed data. They are also rarely operating in isolation. Instead, they draw information from multiple systems, interact with existing infrastructure and become increasingly embedded within day-to-day operations and decision-making.
That matters because many public sector organisations are still struggling with siloed data, inconsistent monitoring tools and limited visibility across hybrid environments. As AI projects scale, maintaining a clear view of systems, applications and data flows becomes far more important, not only for performance and efficiency, but also for governance, resilience and public trust.
This is one reason why observability is becoming a far more important part of the public sector AI conversation. At its simplest, observability is about giving organisations greater visibility into how systems, applications and data are performing across increasingly complex environments.
Without that visibility, maintaining control becomes significantly harder. Issues relating to performance, security, compliance or data quality can become difficult to identify before they begin affecting services or operations.
The focus on resilience
Many IT teams are already using AI to support monitoring, observability and IT operations management. Used effectively, these tools can help organisations identify issues faster, automate routine tasks and improve operational efficiency. But without the right foundations in place, AI also risks creating additional operational pressure, forcing teams to validate outputs, double-check recommendations and manage growing volumes of alerts and insights.
This is also why resilience is becoming increasingly important. Public sector organisations are under constant pressure to maintain service continuity, protect sensitive information and ensure critical systems remain available and secure.. AI may help improve efficiency and automate operations, but it also increases dependency on interconnected systems and reliable data flows.
Indeed, IDC warns that public sector organisations that fail to modernise operational monitoring and observability risk falling into a “vicious cycle” of higher operational costs, outdated operational practices, increased security and compliance risks, and more frequent downtime.
That focus on resilience is only likely to intensify as governments and regulators place greater emphasis on cyber security, operational continuity and digital trust. Legislation such as the Cyber Security and Resilience Bill reflects the growing expectation that organisations must be able to identify, manage and respond to risks across increasingly connected digital environments.
All of this is important because the government is investing heavily in AI. The AI Opportunities Action Plan sets out how it is moving from a period of ‘ambition to delivery’.
In its latest update, it reports that it has now met its commitments against 38 of the 50 actions and has a public dashboard to mark its progress.
For instance, it highlights how it now uses an AI-powered scribe for government meetings to support 1000 officials in 22 local authorities, while another tool is designed to speed up the planning process. While in the NHS, one-third of chest X-rays are now AI-enabled.
Building AI systems that can scale
While people may focus on the end product, the government also makes it clear that “compute is the essential foundation for modern AI” because it is the “processing capacity that enables models to be trained, tested and deployed at scale”.
No doubt all of this will continue to be scrutinised closely by the NAO as AI becomes more deeply embedded into public services and operational decision-making across government. And rightly so.
As the public sector shifts from a period of experimentation to AI adoption, the challenge is to ensure that organisations have the infrastructure, visibility, governance and resilience needed to support those ambitions in practice.
For public sector organisations, this means the focus must now shift from experimentation to operational readiness. Those looking to generate meaningful ROI from AI investments should prioritise strengthening data quality, improving visibility across hybrid environments and modernising operational monitoring and resilience capabilities. Without those foundations in place, even the most promising AI initiatives risk struggling to scale or deliver long-term value.
Rich Giblin
Rich Giblin is Head of Public Sector and Defence at SolarWinds.
