“Sluggish technology has hampered our public services for too long, and it’s costing us all a fortune in time and money,” said The Rt Hon Peter Kyle MP (Secretary of State for Science, Innovation and Technology) announcing a shake-up for the public sector in January this year.

All of us can testify to the sluggishness, whether it concerns (lack of) communication among medical staff caring for us or the challenges of dealing with the local council.

Will AI live up to hopes that it can transform productivity in public services – and even better, help improve services while cutting costs?

AI applications certainly have tremendous potential to save time and money, and improve service quality, in areas ranging from health and adult social care to criminal justice and local government. But it is often difficult to distinguish hype from realistic opportunities, while unlocking those opportunities will require different ways of working in the public sector.

In recent research at the Bennett Institute for Public Policy, we conducted several workshops in collaboration with Professor Wendy Hall, Director of the Web Science Institute at the University of Southampton, bringing together academics and practitioners to explore the opportunities and barriers to AI adoption in public services. Additionally, with Professor Bart Van Ark, Managing Director of The Productivity Institute, at the University of Manchester, we have specifically looked at the responsible use of AI in adult social care. (Both of these studies will be published soon.)

Achieving its potential

Our work, along with that of other researchers, reveals several key themes. One of the most prominent is the central role data plays in AI adoption.

The process of organising data into a consistent and usable format, overcoming barriers to linking datasets, addressing privacy and security needs, and gaining access to comprehensive data for service improvements are all time-intensive tasks.

The government has announced plans for a National Data Library but the need for painstaking construction of high quality and accessible datasets also needs less exciting interventions such as clear regulations and setting common standards for the public sector, as discussed in a new policy brief by Dr Jeni Tennison.

Another broad theme is the shortage of AI know-how in public services. While there are individuals with deep expertise, a broader understanding of how to effectively use AI and its policy implications is not widespread.

This is one reason why – when the Bennett School of Public Policy opens in Autumn 2025 – we will launch a new MPhil programme in Digital Policy for mid-career officials and analysts to contribute to broadening the base of know-how about AI in the public sector.

Finally, changes will be necessary to update governance and accountability structures in response to these rapidly evolving times. AI is an information machine that provides users with predictions or suggested decisions. The productivity benefits will rely on employees' ability to competently use the information in a skilled way and feel empowered to make informed decisions.

There are many pilot schemes across the public sector that suggest tremendous potential; for example one council claims using AI in adult social care has led to 44% administrative time savings for its workers, enabling it to start reducing a backlog of need.

But scaling up AI use across the public sector presents a significant challenge – one that our multidisciplinary research at the University of Cambridge, in collaboration with other academics, policymakers, business leaders and public sector workers, aims to address.

Dame Diane Coyle is Bennett Professor of Public Policy and Co-Director of the Bennett Institute for Public Policy. The Bennett School for Public Policy will open in October 2025, established with the support of the Peter Bennett Foundation. It will offer the well-established MPhil in Public Policy, the new MPhil in Digital Policy, and host the Bennett Institute for Public Policy research centre.

Published: 27 March 2025

^{The text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License}