The recent Mandarin/Liquid survey of public service attitudes to AI showed public servants are more optimistic about AI than the general public, with 84% saying they feel personally ready to use it in their work. Yet this enthusiasm isn’t matched by organisational readiness, with fewer than half believing their agency is equipped to support adoption.

Public service enthusiasm for AI isn’t about grand transformations; rather practical, problem-driven solutions that reduce administrative burden, streamline workflows, and improve navigation and responsiveness.

But for most, current use of AI is minimal, fragmented, and typically limited to generic chatbots and copilots rather than fit-for-purpose tools that address core service challenges. And while some governance frameworks exist, they are not yet supporting on-the-ground AI usage.

The survey shows that the time for action is now. Without good on-the-ground leadership, AI adoption in government will be significantly set back – either through frustration and disillusionment quashing the current enthusiasm, or growing levels of unauthorised use introducing new unmanaged risk. But with good leadership, there are many tangible and realisable benefits that can help deliver better public services.

Public sector leaders need to create the bridge between individual readiness and organisational capability by activating safe experimentation, championing real use cases and modelling responsible adoption.

From hesitation to momentum: How government leaders can enable safe AI adoption
Leadership hesitancy is understandable. The public sector has a tradition of risk aversion that discourages experimentation. No-one wants their project to be the subject of audit scrutiny or reputational damage. Against that default setting, AI is complex, technical, hard to understand, and its inherent uncertainty ratchets up the risk level. Most leaders lack confidence in their own AI knowledge or ability to judge what is or isn’t appropriate, and few have ready examples of successful AI projects on hand to learn from.

With the latest statistics from MIT’s Gen AI report showing 95% of AI projects fail, there’s even less inclination to champion AI adoption. Unpacking why there’s such a high failure rate, however, shows the way forward and highlights the need for better leadership:

Know what problem you are trying to solve: Most AI projects are poorly defined. Giving an organisation a generic AI tool like Copilot or ChatGPT and hoping that staff find a use for it is not a recipe for success. Always start by defining the problem, not the technology.
Choose the right tool for the job: AI contains a broad suite of very different tools and technologies. Find out which type of AI is best suited to the problem you are trying to solve.
Not having the data needed: Organisations are awash with data, but AI projects often fail because they don’t have the right type, quality or quantity of data. Map the data needed before you start, and allow enough time to gather or create it.
It doesn’t fit into the flow: Even when AI projects work, they can struggle to survive because they’re disjointed from the established workflow. Always consider how you can integrate the AI seamlessly into existing processes rather than adding further fragmentation.
Lack of leadership support to scale: Even when initial trials succeed, a lack of leadership support, governance pathway, and organisational direction and priority often means the project can’t scale to become operationalised. As a result, it fails to deliver the anticipated benefits.
Despite this, AI is being successfully used in government today. Early adopters in primary industries, finance, defence and public safety all share two common traits – lots of data and an incremental approach. Their AI projects started small and grew incrementally, solving one problem at a time and embedding it into BAU before moving on to the next opportunity. Then, they scaled carefully, delivering benefits and momentum with each step.

Finding visible success stories to learn from is hard. Communities of practice and technical working groups do exist, but those are aimed at technical practitioners. What’s needed is leadership-level collaboration to build knowledge and confidence on the safe way forward.

Leaders need to become enablers for AI adoption, and develop governance as a safety net — not a blocker
Public sector leaders don’t need to become AI experts, but they do need a basic level of literacy of AI and new governance skills to manage the new paradigm.

We need to reset the structural caution that is holding back the enthusiasm we saw in The Mandarin/Liquid survey. This means addressing uncertainty about the technology, concerns about reputational risk, and lack of clarity on how to balance opportunity with oversight that exists across most leaders and managers.

This starts with all leaders developing a working vocabulary of AI risks and opportunities, and openly using AI personally for routine tasks. This normalises AI as an accepted productivity tool and encourages an open learning culture that legitimises safe experimentation by making it clear that trying, learning and iterating with AI is encouraged within defined boundaries.

In collaboration with other departmental leaders, consider creating an AI playbook that gives practical, on-the-ground guidelines, checklists and templates for AI projects. The Mandarin/Liquid survey identified that current national frameworks provide valuable scaffolding and principles, but not the on-the-ground permission and advice employees need to activate AI projects.

The safest way to start is by fixing the friction staff already know too well
Next, leaders should invest in visible, low-risk/high-friction, problem-driven AI projects that reduce administrative burden, such as automated document and form handling, case triage, and service navigation. These were the areas that respondents to The Mandarin/Liquid survey were already identifying as good opportunities. These initial projects will demonstrate impact and prove AI’s value in a safe, lower-risk internal space, building momentum, knowledge and confidence.

To succeed, leaders need to provide active and enabling sponsorship; establish the priority, create the space, ask the right questions, understand the risks, and seek advice when needed. Actively supporting those public servants who are ready to lead AI projects (54% of survey respondents) will unleash real momentum and improvement.

Make governance a welcome safety net, not a handbrake
Starting small, proving the benefits, embedding, and growing incrementally also provides leaders the space to develop the right style of enabling governance, providing guardrails and a safety net for AI projects.

This includes consideration of risk-tiered processes that fast-track low-risk AI applications, provide simple assurance checklists that staff can use without specialist training, and strengthen accountability through designated skilled AI stewards. The Singaporean government, for example, holds AI clinics where managers can bring their problems to get advice from AI experts to set them in the right direction.

AI literacy and governance urgently need to become core skills for all government leaders
AI capability is a workforce infrastructure issue for government. Developing new technical skills alone won’t enable adoption. Human-in-the-loop oversight is essential for the responsible use of AI, but we also need manager-in-the-loop leadership as a foundation for responsible development of AI.

Leaders across all departments are the key to safe, successful adoption. Supporting their professional development is the first step to tapping into the enthusiasm across the organisation and realising the benefits of AI safely.

Mind the gap: Why government leaders must become better enablers of AI and how to do it

Sign up to our newsletter.