Our public service is made up of people who care deeply about doing the right thing. Approaching innovation and AI in a way that is anchored in pragmatism and responsibility sets up projects, people and the public for future success. Most importantly trust is nurtured, not eroded, which is a significant risk when addressing new technologies.
Our recent survey of senior public sector executives* shows that AI adoption is underway across government. The story is not one of fear or resistance, but of commitment to capability building and public value. Government is leaning into change, looking for clear pathways and practical support to make innovation safe, scalable and effective.
Agencies are experimenting, frameworks are forming, and capability is growing. The opportunity now is to connect and scale these efforts. With the right structures and support, government can move from cautious exploration to confident delivery.
What we found
- Activity is high: most agencies are already piloting or using AI.
- Scale is low: only a small number have moved beyond pilots into embedded use.
- Confidence is mixed: many leaders feel only moderately confident about responsible adoption.
- Barriers are structural: skills and trust are the top constraints, well ahead of funding or technology.
- Where the value is: leaders see the biggest opportunities in internal efficiency and service delivery
The takeaway is that this is not about appetite or technology. It is about confidence, coordination and the right operating model.
The stakes
| Not getting it right | Identifying and unlocking value | |
| Without coordination, AI adoption risks becoming a patchwork of isolated pilots and projects, with duplicated investments and efforts, and inconsistent governance. This will lead to slower progress and an erosion of public trust. | The opportunity is far greater. With the right leadership, focus on value and structure, AI can improve productivity, enhance service delivery and build stronger public confidence in government’s ability to innovate safely. |
The optimum AI operating model – a balanced approach
The question is no longer whether government should use AI, but how it should be governed and supported to deliver the greatest value.
Too much central control can make implementation rigid and disconnected from real-world needs, leading to fragmentation, and hidden risk. While too much local autonomy can lead to duplication, inconsistent standards and higher delivery risk and costs.
Our analysis suggests a Balanced Centre model that combines central guardrails with local delivery and innovation works best.
Options and trade-offs
| Model | What it is | Strengths | Challenges |
| Centre-led | Central team sets policy, approves use, builds common platforms and services. | Consistency, visibility, unified standards, easier to attract scarce talent. | Hard to localise to department context; can slow decisions; risk of “shadow AI” if processes feel blocking. |
| Department or agency-led | Each department decides, procures and runs its own solutions. | Fast, mission-aligned, responsive to local needs. | Duplicated effort, fragmented data and controls, inconsistent guardrails, higher total cost. |
| Balanced Centre (recommended) | Centre sets policy, standards and shared services; delivery happens in agencies within guardrails. | Clear rules, shared learning and reuse, local ownership, faster value with controlled risk. | Requires active coordination, clear roles, and ongoing sponsorship to stay enabling rather than controlling. |
This approach reflects what public sector leaders told us they need: trust, collaboration and practical support, not one-size-fits-all control.
A perspective across jurisdictions
States are already making progress. Departments are experimenting, learning and building capability. The opportunity now is to connect efforts through a shared operating model that accelerates what is working, reduces duplication, and builds confidence across the system.
By linking clear guardrails with practical support, Governments can move from pilots to lasting value. These examples illustrate how other jurisdictions have balanced central leadership with local delivery.
| Jurisdiction / Organisation | What they did | Model signal | The key takeaway | |||
| United Kingdom | Central principles and playbooks; departments implement in context; shared enablers for common needs. | Balanced Centre | Plain-English guidance, reuse of code/patterns, clear roles between centre and departments. | |||
| Canada | Central AI Centre of Expertise and shared infrastructure; departments run use cases aligned to a federal strategy. | Balanced Centre | Central advice and assurance patterns; departmental accelerators; communities of practice. | |||
| New Zealand | System-wide AI principles and a public service AI framework; agencies retain autonomy to implement. | Balanced Centre | Principle-led governance, voluntary alignment, high transparency and sharing. | |||
| Singapore (GovTech) | Common stacks and platforms delivered centrally with hands-on enablement for agencies. | Centre-led with enablement | Small expert teams, common building blocks, strong service enablement instead of heavy gatekeeping. | |||
| Estonia | Shared digital backbone and standards across agencies; agencies build on top. | Centre-led | Invest in common plumbing and interoperability; keep delivery close to users. | |||
| Large enterprises (e.g. major banks, multi-nationals) | Enterprise standards and risk controls centrally; business units and regions deliver use cases. | Balanced Centre | Central risk and data controls; federated delivery aligned to business value. |
Real world application
To move from pilots to real impact, governments can:
- Set the tone from the top – Ministers and Secretaries must set the tone for how AI delivers public value. Defining desired outcomes, transparency, and safety. Leadership should focus on benefits and responsible use, not specific tools. While they don’t need to be technical experts, they do need a strong foundation in AI literacy to ask the right questions, understand risks, and make informed, value-based decisions that guide projects toward meaningful and accountable outcomes.
- Design for delivery – Translate frameworks into operating models that make responsible action easier: who decides what, where assurance sits, and how reuse happens.
- Show value early – Start with safe, high-friction use cases, such as document handling or workflow triage, to build trust and momentum. Drive coordination from the centre to publish playbooks and designs others can reuse.
- Build platforms that agencies want to use – Make them modular, reliable and easy to access. Adoption is earned through reliability, speed and fit.
- Sustain capability and investment – Create a small enablement team that embeds within agencies to lift skills and confidence. Fund training, change and communities of practice. Track outcomes, not just launches.
Why the Balanced Centre drives value
- Confidence and safety: common guardrails and simple assurance patterns reduce risk without stalling delivery.
- Speed with control: departments deliver in context; the centre supports reuse and clears blockers.
- Better value for money: shared patterns, components and procurement reduce duplication and improve efficiency.
- Learning that compounds: visible wins and reusable playbooks turn isolated pilots into system-wide progress.
A shared opportunity
Public servants are already taking thoughtful steps into AI. They are experimenting, learning and testing what works. Their caution reflects care for the people they serve, not resistance to change. By connecting clear guardrails, practical tools and shared capability, AI can enhance the value of public service and deliver better outcomes for citizens.
About the research
These insights are drawn from KordaMentha’s AI in the Victorian Public Sector survey and panel discussion with government, academia and industry leaders. Our focus is helping government turn intent into delivery through practical operating models and reusable playbooks that make responsible adoption easier
*Based on a survey of 31 senior Victorian public sector executives conducted by KordaMentha in August 2025
