AI adoption in the UK public sector is accelerating rapidly, with many organisations investing heavily in AI in public sector services and operations to reduce manual effort, improve decision-making and stretch limited capacity further.
In many cases, the technology performs well in pilot environments. Challenges emerge when organisations attempt to scale it into live operations, where familiar issues begin to surface. Manual intervention remains high, exceptions continue to require human judgement, governance requirements increase, and the efficiency gains that justified the investment remain out of reach.
In our latest whitepaper, “The New Shape of Public Sector Transformation”, we examine why many transformation efforts stall at this point and struggle to deliver the outcomes expected.
Reinvigoration’s work with public sector leaders, alongside evidence from the NAO, Socitm and the Central Digital and Data Office, points to a consistent conclusion: AI cannot deliver its full value in operating environments that were not designed to support it.
The NAO’s 2024 review of government digital transformation highlights the same issue. Individual digital services have improved, but the underlying operations, data and legacy technology required to support them have not kept pace. Departments have focused on improving the visible, public-facing layer, while the operational infrastructure beneath it continues to struggle.
This is not a new dynamic. Previous waves of public sector technology investment have followed a similar path, improving access and experience at the front door without addressing how services are actually delivered behind the scenes.
At the same time, the nature of work itself is shifting. Increasingly, activity is driven by complex, multi-agency casework rather than standardised transactions, requiring greater professional judgement and coordination across organisational boundaries.
We explore this broader challenge in more detail in our blog, “The Problem With Public Sector Transformation.”
Where services are delivered in different ways across teams, locations or systems, attempts to scale AI in public sector operations begin to expose those differences. Variations in data, decision-making and process design that were previously managed informally become more visible as organisations try to standardise them.
This is often the point at which transformation efforts begin to lose momentum.
This aligns with analysis from The Productivity Institute, which highlights how productivity gains from technology are frequently constrained by underlying operational complexity rather than the technology itself.
A common assumption is that introducing AI into operations will smooth out complexity. In practice, it often carries that complexity forward.
This is particularly visible where AI is used to automate elements of a process.
Where a service operates in multiple ways depending on team, system or location, automation does not remove that variation. It replicates it.
Instead of simplifying how work is done, organisations end up with multiple automated pathways, each requiring oversight, maintenance and explanation. The expected efficiency gains are diluted, and the burden shifts toward managing that complexity in a different form.
This is one of the reasons AI struggles to scale in fragmented operating environments.
Recent challenges in NHS technology programmes reflect this dynamic. Delays in electronic patient record implementation, interoperability issues between trusts, and systems that do not communicate across organisational boundaries are often framed as technology issues. In practice, they are closely linked to the complexity of the operational environments those systems are expected to support.
The 2026 Interim Report of the National Maternity and Neonatal Investigation highlighted how fragmented IT systems were adding pressure for frontline staff.
Patient information was spread across multiple systems and, in some cases, duplicated in paper records, requiring staff to transfer information manually between systems while continuing to deliver care.
In critical situations, clinicians were navigating several systems at once, none of which were fully integrated with each other or with wider hospital systems. This is where operational complexity becomes visible.
Digital tools have been introduced, but the underlying operating model has not been simplified, with the result that technology adds effort rather than reducing it.
This dynamic is not unique to maternity services and can be seen across the public sector wherever complex operations meet digital ambition.
The regulatory context adds further complexity.
There is currently no single AI regulator in the UK, with oversight sitting across multiple bodies including Ofcom and the CQC, whose frameworks were not designed with AI systems in mind. For organisations handling sensitive data, this creates additional challenges in deploying AI responsibly and demonstrating how decisions are made under scrutiny.
At the same time, regulatory approaches are shifting toward outcomes rather than process. Frameworks such as the CQC’s single assessment approach, Ofsted’s inspection methodology and the Regulator of Social Housing’s consumer standards require organisations to demonstrate consistent outcomes, not simply that the right processes are in place.
Where ways of working are inconsistent, introducing AI makes that consistency harder to evidence. In practice, this often leads to additional oversight and manual intervention as organisations work to meet regulatory expectations.
For AI to scale across services and how they are delivered, the wider operating model needs to be consistent.
This means clear end-to-end ownership of processes, defined decision logic and governance for automation, aligned data definitions across teams and systems, and clear accountability for exceptions and edge cases.
Where these conditions are not in place, organisations tend to compensate by adding controls. Additional review steps are introduced and oversight expands, meaning that while the technology may function as intended, the overall operation remains difficult to manage and change.
The “digital front, analogue back” pattern is already well established. Socitm Digital found that while 72% of councils had invested in digital platforms, fewer than 25% had integrated them with back-office operations.
72% of councils have invested in digital platforms, fewer than 25% have integrated them with back-office operations
This gap between the front door and underlying operations has been a recurring feature of public sector transformation, and AI in public sector delivery risks reinforcing it if the operating model remains unchanged.
Without addressing these underlying issues, AI is likely to follow the same path as previous transformation efforts.
The organisations that will benefit most from AI will be those that prepare their operations to support it.
This starts with understanding how work actually flows in practice, making it visible end to end, reducing unnecessary variation and clarifying ownership before introducing new technology.
The CDDO estimates that around 28% of the government’s most critical systems require remediation, which creates a challenging foundation for scaling AI effectively.
Simplification creates the conditions for AI to deliver value at scale.
The question is no longer whether to adopt AI. It is whether operations are simple, consistent and clear enough to make it work.
Our Simplify4Scale® methodology supports this by providing an evidence-based view of how work flows through your organisation before technology investment is made. By addressing complexity first, organisations improve the likelihood that AI delivers sustainable value.