Earlier this year, tech industry leaders blasted the UK Government for offering just £80k for a senior role in its new ‘sovereign AI’ unit, branding the salary “laughable” given the scale and ambition of the brief. At a time when AI adoption in government is cranking up – with the National AI Strategy and the AI Opportunities Action Plan pushing things forward – it signals a real mismatch between ambition and investment.

And it’s not just in the UK. In the US, the Department of Government Efficiency (DOGE) is pursuing an AI‑first strategy to remake the government, including an “Doge AI Deregulation Decision Tool” that aims to delete regulations it deems as no longer required by law.  

However, technologists – who are in fact pro-government automation – have raised concerns that such technologies may not be sufficiently capable and may instead increase inefficiencies. In fact, a Washington Post exposé highlights a number of recent examples where relying on automation alone to make critical decisions has led to big government mistakes. In New York City, for example, an official AI chatbot last year advised businesses to break the law. And then there was Australia’s infamous Robodebt scheme – a flawed algorithm that wrongly reclaimed benefits and ultimately cost the government over a billion dollars in compensation.

So, it begs the question, are we focusing on the right problems? And more fundamentally, do we even need AI for everything?

Too often, departments are defaulting to AI when simpler, safer, and more cost-effective technologies could do the job just as well, or better. What’s more, if we’re not willing to fund the expertise needed to lead AI responsibly, then perhaps we shouldn’t be using it at all.

With this in mind, I believe every senior stakeholder can benefit from a clear AI playbook; one that asks the hard questions from the outset.

Are you taking a business and problem-first approach? Not everything needs to be automated, and not all automation requires AI, so how do you separate hype from value and establish if you actually need AI? First, AI initiatives must align with the core business strategy and demonstrate clear public value delivery. Avoid rushing implementation or over-investing in unproven solutions, and instead, focus on organisational problems rather than seeking AI use cases. This ensures AI implementation addresses genuine business needs rather than pursuing technology for its own sake.

Consider strategy integration and ownership. Rather than existing in isolation, AI strategies should be embedded within broader organisational frameworks – such as within a comprehensive tech and data Strategy – and require overarching coordination teams that work across all organisational areas. This integration ensures AI development aligns with existing technology investments and data management initiatives, while ensuring consistent approaches, shared learning, and avoiding duplicated effort or conflicting implementations.

Take a phased development approach. Start with high-level strategic frameworks and evolve them through practical experience rather than attempting comprehensive solutions before initial release. This approach enables organisational learning while maintaining momentum toward operational implementation.

Lay strong data foundations. Before implementing AI, departments must prioritise data maturity, architecture, and quality. From our experience, around 80% of AI effort lies in data preparation – and many departments begin from a low baseline. Legacy systems and siloed data structures limit accessibility, so modernising data architecture (e.g. through data lakes or mesh frameworks) is essential.

Establish solid governance frameworks. Effective AI governance requires clear policies and guardrails that help users understand acceptable use boundaries. Users demonstrate greater confidence when they understand what is permitted and prohibited.

Develop comprehensive risk registers and conduct thorough risk analysis. The balance between risk mitigation and opportunity capture can be a challenge, but it’s vital, requiring systematic identification of potential harms and development of appropriate mitigation strategies.

Build trust through transparency. Trust in AI hinges on transparency and explainability, meaning government departments should demand clear, understandable reasoning behind any AI-generated recommendations – especially in high-stakes areas like healthcare, where human oversight remains essential. Addressing public concerns about fake news, fraud, and AI-generated content quality is also key. Here, clear communication strategies can help build public understanding and trust, while maintaining appropriate transparency about capabilities and limitations.

Don’t underestimate the importance of vendor management. As well as asking for reasoning behind recommendations, demand clarity – contracts should include explainable outputs, transparent audit trails, and compatibility with legacy systems, so that critical data isn’t left behind. In addition to this, to avoid lock-in and ensure long-term flexibility, I recommend departments prioritise vendor-neutral approaches and consult with other government agencies or SME consultancies to share learning and strengthen bargaining power.

Evolve, continuously. Know that what works today, may not next month, so look to develop iterative, flexible strategies that evolve with experience, organisational learning and technology maturity.

For me, governments’ biggest risk isn’t a lack of AI ambition – it’s a lack of alignment and investment in the people expected to deliver it. The strategies above show that effective AI adoption isn’t just about technology; it depends on everything from strong data foundations, governance, and iterative development, all the way through to trust, transparent processes and people.

There’s no doubt that the pace of policy development is rapid, but without matching investment in senior leadership, capability, and delivery teams, even the most well-designed strategy is at risk of falling flat.