At the 2024 Digital Leaders National Digital Conference in October, I had the opportunity to speak with public sector innovators Dr. Laura Gilbert CBE from No.1 Downing Street’s AI Accelerator and Simon King, Director of AI at the Department for Work and Pensions, and Professor Mark Thompson, about the safe, ethical, and accelerated scaling of AI within the UK Public Sector.

Having worked with government and private sector partners for almost two decades to deliver data-led digital transformation at scale, I’ve been privileged to witness firsthand in recent years, through our work with the NHS and NatureScot how AI can be used to deliver the next generation of national public service transformation.

Yet, while these examples are significant, the real question remains: how can we delivery wholesale transformation through AI and how do we tackle the barriers to AI-enabled transformation in the public sector?

The State of the AI Nation in Public Services

On one hand, the potential benefits of AI for public service enhancements are becoming increasingly recognised by civil servants and innovators in the public sector. A March 2024 report by the Alan Turing Institute identified approximately 143 million government transactions that AI could streamline, potentially saving 1,200 person-years’ effort annually.

Furthermore, the newly established Regulation Innovation Office seeks to fast-track the integration of emerging technologies, including AI, into public services by reducing regulatory hurdles – signaling the appetite and support for AI-enabled transformation and innovation across government.

Yet examples of scaled AI deployment in government remain scarce. For instance, a 2024 Digital Leaders survey revealed that many AI experiments struggle to transition beyond initial stages, with a low return on investment for AI projects across the board. Similarly, the National Audit Office’s paper on AI in government found that, while 70% of public sector leaders are piloting AI use cases, only a small fraction advance to widespread deployment. In many cases, we hear more about proofs of concept and AI widgets rather than about successful, fully operational solutions delivering real, measurable impact.

There was clear consensus across the innovators in the room at the National Digital Conference that to move from point-based technical proofs of concepts to scalable, AI-enabled transformation, we need to put users front and centre, focussing on trust and buy-in with far greater consideration of organisational transformation and leadership.

Three Areas to Focus Action for Scaling AI Services in Organisations

Through our own experience of scaling AI deployments, we’ve identified three critical areas that require focused attention to bridge the gap between pilot projects and live solutions:

1. Doubling-Down on a User-Centric Approach

In the rush toward AI solutions, many teams overlook critical user-centred practices that have been proven to deliver better outcomes. AI, as a high-profile, disruptive, and often misunderstood technology, requires an even greater focus on real user needs so that it doesn’t become purely technology-led.  This means involving both citizens and government staff in the design of AI-enabled services, ensuring users understand, trust, and feel a sense of ownership and agency over new solutions and what it means for the way they work.  Moreover, rapid iteration of designs with users can help test real-world feasibility early, avoiding costly investments that simply don’t have a practical ‘path to live’.

2. Data-Centric Approach: AI Readiness?

AI solutions are only as good as the data on which they rely. This means that data must be available, reliable and critically, fit for purpose for the intended AI techniques. Too often, the rush to experiment with AI skips over essential steps like ensuring data privacy, obtaining consent, and assessing the quality and completeness of data for use at scale.

Before we start building AI solutions, a rigorous assessment of available data—its provenance, suitability, and limitations—should be a priority. By identifying potential data challenges early, teams can design services that are transparent and realistic in what they promise. This early focus on data fitness-for-purpose helps mitigate the risks of delayed or ineffective AI deployment, especially when deployed at scale.

3. Governing AI-Specific Risks

AI introduces a plethora of new challenges including areas such as explainability, bias, and data privacy. These are no longer abstract challenges but critical barriers with respect to effective, safe and responsible deployment of AI in live environments. Without a proactive approach to identifying these issues and how they can be practically addressed early, even the best technical solution may never be suitable for live operation.

Effective AI service design requires close and transparent engagement with stakeholders in areas such legal, policy and compliance, helping them to understand risks, impacts and potential mitigations. This includes using robust frameworks for explainability to ensure AI decisions are understandable, addressing biases that may be present in training data, and establishing strict controls for data privacy and consent. In many cases, a delay or failure to ask the right questions about these factors upfront can stall an otherwise promising initiative.

Moving AI from Pilot to Scale: A Path Forward

To truly bring AI to scale in public services, we need to evolve how we approach discovery and service design. This means integrating human-centred practices that have historically driven digital success, while sharpening our focus on data readiness and risk management. Responsible AI innovation requires transparent, collaborative engagement between government and suppliers.

If we get these elements right, the potential for AI to transform public services is vast. However, realising this potential requires an understanding of AI’s unique challenges and a commitment to user-centred, data-driven, and strategic governance practices. Only then can AI projects move from small-scale experiments to fully-fledged solutions that transform public services at a national scale.