Turning AI ideas into scalable public services: Lessons from three years of the AI Challenge

The Futurescot AI Challenge is an annual competition that brings together Scottish public sector organisations to solve real-world operational problems using artificial intelligence. Conceived by Futurescot media Ltd and StormID, the initiative helps public bodies turn raw AI concepts into practical, scalable proofs of concept (PoCs)

Those lessons matter more than ever. Scotland's Public Service Reform Strategy commits public services to being preventative, joined up and efficient, and names AI as part of the answer, from predicting demand to scaling automation. Within the Scottish Government itself, the Serving Scotland five-year plan aims for AI and automation to become part of everyday work by 2030 and Scotland’s AI strategy states that Scotland must innovate in the adoption of AI within public services

With AI tools now widely available and expectations rising, the question is therefore shifting from whether public services should use AI to where it creates enough value to justify the investment, risk and organisational change involved. Drawing on our assessments of applications, this post sets out what makes an AI idea worth pursuing and the common challenges organisations must resolve to realise the benefit.

Five tests we now apply to public sector AI ideas

Drawn from over 80 applications, these are the questions that most often separate ideas that progress from those that stall.

1.Does it need AI? Many needs can often be met by off-the-shelf products or conventional software, such as a well-designed online form. A chatbot request often points to a content problem: guidance that is scattered, out of date or hard to find. Fix the content and the case for a chatbot often weakens. Similarly, requests for AI-generated insight across separate systems and spreadsheets usually need data integration first. After that, good BI and analytics may already meet a substantial part of the need.

2. Do you understand your business process well enough? Most ideas we saw applied AI to individual steps within an existing process. That's a sensible starting point: quicker, lower risk and a good way to build confidence. It works best where the process is already written down, such as a referral pathway, assessment criteria or statutory guidance. Where the rules live in people's heads, discovery needs to come first. Mapping the end-to-end workflow also shows where integration with case management systems may present challenges. Just as importantly, it reveals which steps exist only because of how things have always been done, which is the starting point for the more radical redesign we argue for below.

3. Can you evaluate the use case? Even strong ideas need plenty of realistic, anonymised examples and a human-judged baseline to measure the AI against. Teams routinely underestimate how long this takes, especially where consent and privacy are involved.

4. Is there a clear path to live deployment? Privacy need not always prevent early experimentation, as proofs of concept can often use anonymised or synthetic data. The harder questions come when moving beyond it: consent, data processing agreements, security approvals and where the AI is allowed to run. Many high-value services, for example in health, policing and justice, would benefit from private AI infrastructure.

5. Do you have the skills and data to make it work? Many organisations want to use AI for prevention, particularly in education and health. That needs data science skills and model training: a very valid ambition, but a far longer journey than deploying a ready-made large language model from a cloud provider.

Five use cases we see working most consistently

From across 80+ applications these are the AI Challenge use cases which have consistently demonstrated both strong value and feasibility.

1. Triage and assessment. This is the strongest and most reliably feasible pattern. AI reads unstructured input such as complaints, enquiries, clinical referrals, CVs and application forms. It then extracts key details, classifies them and checks them against criteria, provided the process is clearly understood. It is also one of the repeatable patterns identified in our 2026 white paper, Automate tasks, not jobs: The AI opportunity for Scotland's public services.

2. Understanding legacy systems. Using AI to analyse ageing, poorly documented applications, to decide whether to keep, refactor or replace them, is both high value and highly feasible. Public bodies are starting to recognise this.

3. Knowledge assistants. Staff spend a lot of time hunting for the right policy, procedure or past decision, or finding themes in unstructured data. Assistants that let inspectors, caseworkers, advisers and frontline teams query their organisation's own guidance save time where it's felt most, provided good, well-categorised content sits underneath.

4. Document capture and creation. Demand for speech-to-text and summarisation has grown every year, for example in health, social work and local authorities and there are many available tools which can support this area.

5. Predictive early intervention. This is a common need, and the Public Service Reform Strategy names predicting demand as an AI priority. That makes the data foundations, and building specialist AI talent in Scotland, even more important.

The bigger opportunity: what needs to happen next

Most applications to the AI Challenge have applied AI to one organisation problem within an otherwise unchanged process. That's a sensible start, but on its own it won't deliver the joined-up, efficient services national reform demands. We think this needs a step change in approach.

1. From system-led to service-led design. Start with the outcome a service needs, then ask how it would be designed if AI were available from day one. Bolting AI onto today's process risks automating steps that shouldn't exist. Many of those steps exist only because legacy line-of-business systems forced them: for years, processes have been shaped around what the system could do. AI now makes analysing and rewriting those systems far quicker and cheaper, so the service design can lead and the system can follow.

2. From silos to joined up user journeys. People experience public services through life events that span councils, health boards and national bodies. The Public Service Reform Strategy shows where this needs to go: AI that works across those boundaries, built on shared data and common standards.

3. From pilots to platforms. Design each successful AI project as a reusable component from the start, Scotland's size should make "build once, use many times" genuinely achievable. We've seen first-hand the use cases across many organisations are largely similar.

4. From assistants to accountable agents. As AI capability and trust in it matures, AI can move from drafting to acting. That needs clear rules, permissions and audit trails, and deliberate choices about what should always stay human. Estonia is working towards an "agentic" state in which AI agents can act on behalf of citizens and officials.  That depends on a single digital identity, a secure data exchange layer and audit trails showing who accessed what.

5. Control of data. Around half of the highest-value public services we analysed in our Tasks not Jobs white paper would need private AI infrastructure, mostly in health, policing and justice. Scotland is well placed to build it, with surplus renewable energy and a climate that cuts data centre cooling costs. Investing in sovereign AI capacity alongside UK public cloud would let the most sensitive services use AI while keeping full control of their data.

Working in the open

With AI moving this fast and so many organisations facing the same challenges, working in the open has never mattered more. At StormID we are keen to continue to share our lessons learned.

If you're working with AI in Scotland's public sector, we'd love to hear from you, understand what's working and what's getting in the way

Drop us a line at ai@stormid.com or come and speak to us at Digital Scotland and we’ve love to share more about we've learned from the AI Challenge and hear about the challenges you're facing