The GovICT Theatre at DigiGov Expo 2025 was packed for an insightful session exploring real-world AI implementations in the public sector. Brian Glick, Editor in Chief of Computer Weekly, chaired a discussion between Søren Madsen, Chief Consultant at KL (Denmark's Local Government Association), and Alex Jones, Interim Director of the UK Government's Incubator for Artificial Intelligence. The 45-minute session provided a rare opportunity to compare notes between two countries at different stages of their AI journeys, with Denmark offering a particularly compelling case study of nationwide municipal AI adoption.
The session opened with a sobering reality check about the journey from pilot to production. While there's significant activity across both countries, success rates remain challenging one speaker noted that when funding 40 pilot projects, only six progressed to implementation. This highlighted a critical theme: innovation in the public sector naturally involves failure, with success rates as low as 5% being typical. The key, it was emphasised, is to fail fast and move on, though this approach sits uncomfortably with traditional public sector structures that tend to persist with projects simply because timelines and budgets have been committed.
Denmark's approach to knowledge sharing emerged as a standout model. The country has developed a municipal AI map: an online platform where local authorities can submit their AI projects and browse what others are doing. With 224 projects now listed, 132 of which are operational across 44 municipalities, this crowdsourced approach provides both transparency and inspiration. Crucially, the map includes contact details for project managers, enabling direct peer-to-peer learning. Whilst the platform is only available in Danish, the principle offers a template that other nations might consider adopting.
The types of projects gaining traction reveal common priorities. Chatbots – both for citizens and employees; feature heavily, despite survey evidence suggesting they're amongst the least desired AI applications from citizens' perspectives. Other popular implementations include AI for meeting minutes and transcription (with claims of up to 23% time savings in some employment offices), solutions for improving the clarity of official letters to citizens, and more specialist applications like fleet optimisation for electric vehicle transitions and AI sensors to prevent falls amongst elderly people in their homes.
Three projects were highlighted as particularly successful scaling examples. The Fleet Optimizer, which helps municipalities determine optimal vehicle fleet sizes during the transition to electric vehicles, has been adopted by around 20 Danish municipalities. Its success was partly attributed to dealing with car data rather than sensitive citizen information, thus avoiding some of the thornier regulatory challenges. Roboref, a meeting transcription and minutes tool built on generative AI, has demonstrated significant time savings and is now being considered for large-scale deployment. Meanwhile, the Muni chatbot, which started as a traditional bot with 8,500 built-in answers, has evolved to incorporate generative AI and now operates across 41 municipalities, having handled over 300,000 citizen conversations.
Regulation emerged as perhaps the most significant barrier to progress. One Danish municipality reportedly spent two months building an AI solution but ten months navigating the legal requirements. With GDPR already imposing substantial compliance burdens, the addition of the EU AI Act has created what was described as an even more difficult regulatory landscape. The challenge is compounded by inconsistency – one municipality's lawyer might approve a project that another's would reject on data protection grounds. There's now a push for joint regulatory clarifications and standardised risk assessments that municipalities can share, though this remains a work in progress.
The discussion acknowledged that scaling successful pilots requires far more than technical competence. Estimates suggested that technology accounts for just 20% of the challenge, with 80% relating to organisational change. Supporting municipalities through implementation means providing project management resources, helping with impact documentation, and facilitating knowledge transfer between early adopters and those following. Denmark has experimented with "technology partnerships" bringing together 20 municipalities at a time, with dedicated project managers helping smaller groups implement solutions. Critically, co-financing from participating municipalities is now required – ensuring they have "skin in the game" rather than treating central funding as money for experimental projects that might never see production use.
Data infrastructure was identified as another critical scaling requirement. In a survey of 40 AI pilot projects, 79% cited data issues as their largest barrier to success. The technical demands of moving from pilot to production are substantially different, requiring investment in platforms and infrastructure that proof-of-concept work can bypass. The skills mix needed also changes, with legal expertise, user-centred design input, and robust evaluation capabilities all becoming essential.
User adoption challenges were explored candidly. Public sector workers face conflicting pressures – some fear AI will eliminate their jobs, whilst others are frustrated by the slow pace of adoption. The historical parallel with spreadsheets was invoked: accountants initially feared electronic spreadsheets would make them redundant, but instead found their expertise became more valuable as they were freed from manual calculations. The panel suggested AI could follow a similar pattern, automating administrative burden and allowing workers to focus on tasks requiring human judgement and interaction. However, ensuring workers understand how to use these tools effectively remains a significant training challenge.
On the question of technology choices, both countries acknowledged the dominance of Microsoft and OpenAI solutions, with one estimate suggesting 99% of implementations use these platforms despite ambitions for greater digital sovereignty. Denmark has an open-source collaboration called OS2 that has developed around 50 solutions, and there's recognition that Europe needs to build competencies to compete. However, pragmatism often wins – European alternatives haven't yet proven to deliver comparable results. The approach taken by some teams is to use different foundation models for different tasks within the same application, allowing them to leverage each model's strengths whilst maintaining flexibility to swap components as the technology evolves.
Political expectations were described as "sky-high" in both countries. Denmark's Prime Minister has called for five large-scale AI solutions that could reduce public sector staffing by 30,000 full-time equivalents. The UK's AI Opportunities Action Plan, launched in January, positioned the country as aiming to be an AI "maker" not just a "taker", with commitments to a 20-fold increase in compute capacity by 2030. Tools like Extract, which uses AI to digitise planning documents such as tree protection orders, were highlighted as potentially transformative; the project was described as a "planning genome project" that could unlock significant economic growth by speeding up planning applications.
The incubator model adopted by the UK Government Digital Service represents one approach to accelerating adoption. This small, engineering-focused team builds and tests AI products that can benefit the public sector, with successful tools then made available more widely. Recent examples include Consult, which automates analysis of public consultation responses, and Minute, a customisable transcription and note-taking tool that's significantly cheaper than commercial alternatives. In a notable move towards openness, the code for Minute was released as open source on GitHub the week before this session.
The session concluded by identifying what both speakers considered their most promising current projects. Extract's potential to digitise planning data across local authorities was highlighted as having economy-wide impact potential. The meeting transcription and minutes tools whether Denmark's Roboref or the UK's Minute – were identified as offering clear value across the public sector, automating tasks that nobody particularly enjoys whilst freeing up time for more valuable work.
A clear message emerged: whilst the technology is developing rapidly, the real challenges lie in organisational change, regulatory clarity, funding models, user adoption, and cross-sector collaboration. Both countries are learning by doing, with Denmark perhaps slightly ahead in creating systematic mechanisms for sharing knowledge and supporting scaled implementation. The transparency offered by Denmark's AI map and its willingness to acknowledge failures alongside successes provides a model that other nations would do well to study. As one speaker noted, having diverse stakeholders represented at events like this one is valuable, but the real work of collaboration across departments, between levels of government, and internationally remains essential if AI's potential in the public sector is to be realised.
Liuba Pignataro
