Australia is delivering on digital government but AI is a different test

Australia has built one of the world’s most advanced digital government ecosystems, with the OECD’s Digital Government Outlook 2026 ranking Australia second out of 42 countries, behind only Korea, and first in the world for user-driven design. Platforms like myGov are also held up internationally as examples of shared government infrastructure done well.

But that same OECD report contains a more sobering statistic: fewer than three in 10 OECD countries systematically assess whether the AI they have deployed in government is delivering results, and only one in four properly evaluate whether digital projects have met their goals. Australia is unlikely to be the exception here, which points to a deeper challenge facing public sector technology leaders. Pilots and proof of concepts are important, but the real test is translating experimentation into meaningful, scalable outcomes. This is playing out across the Australian public sector right now, with many agencies having explored AI through pilots and trials, from chatbots and document summarisation tools to productivity assistants such as Copilot. The APS Copilot trial found that 64% of managers saw improvements in their teams' efficiency and quality when using these AI technologies. This is a promising outcome, and proof that AI can work well in a government setting; the harder question is what happens next.

Moving from pilots to scale

For many public sector organisations, the conversation is shifting from whether AI can help, to how successful use cases can be scaled. This shift is being driven by tighter budgets, rising citizen expectations and growing pressure to demonstrate return on investment rather than just activity. The public sector is not short of good ideas, and many pilots have shown genuine promise, but the harder task is turning something that worked well in one team into a repeatable model that can be adopted across an agency, or across the sector.

This is a structural challenge, not just a technical one. Risk and assurance processes are being adapted to technologies that are evolving quickly, making it harder to reuse lessons, controls and governance approaches that have already been established elsewhere in government. Funding models are often built around annual appropriation cycles, while AI capability needs room to flex and iterate. Accountability can also become less clear once a system moves from a controlled pilot into live service delivery.

Addressing a fragmented governance model

Part of what has made AI harder to scale is the way governance has historically worked across government. Oversight has largely sat with individual agencies, each setting its own risk appetite and approach to responsible use. This makes sense: agencies understand their own data, systems and constituents better than any central body could. But as AI moves from pilots into core service delivery, agency-led governance needs stronger coordination.

Different approaches to assurance, transparency and accountability can make it harder to reuse what works, compare outcomes and build public confidence across government. Avanade’s research shows why this matters: around a third of public service organisations say trust and ethical uncertainty, rather than the technology itself, are the biggest barriers to using AI to its full potential.

The federal government’s recent decision to establish an Office of AI and develop national AI standards is an important step towards greater consistency. The opportunity now is to turn that whole-of-government direction into practical guidance, shared assurance processes and repeatable practices that agencies can use with confidence.

This is not about removing agency flexibility. Public sector organisations must consider service delivery, legislative compliance, transparency and public trust alongside productivity and efficiency gains. This helps explain why the same Avanade research found public sector leaders expect a more measured return on AI investment: roughly twofold within 12 months, compared with the fourfold return expected in sectors like financial services and retail. The aim should not be to push agencies to move faster at any cost, but to give them clearer guardrails so they can scale AI safely, reduce duplication and maintain public trust.

Treating AI as a whole-of-enterprise capability

Addressing these challenges in the public sector requires more than a new framework. It also requires a shift in mindset, from treating AI as a series of standalone projects to treating it as a capability that belongs to the whole organisation.

In practice, that means four things: shared governance that agencies can draw on rather than reinvent each time; common assurance frameworks so a model validated in one department does not need to be re-litigated from scratch in another; reusable data foundations, because many AI failures in government begin with data that was never built to be shared; and clear accountability structures, so there is no ambiguity about who is responsible when a system moves into live service delivery.

Underpinning all of this needs to be a human in the lead, not just a human somewhere in the loop. AI can be extremely effective at drafting, summarising, flagging and predicting, but it should not be making decisions that affect someone’s welfare payment, their visa status, or their access to care. The technology should support and accelerate human judgement rather than replace it, and governance that assumes otherwise risks eroding the trust it’s meant to protect.

What success actually looks like

The Australian public sector has already shown, by the OECD’s own measure, that it can digitise services well. Consistent standards, shared platforms and a genuine focus on user experience did not happen by accident: they reflect years of sustained leadership, investment and disciplined execution.

AI now needs that same level of discipline applied to a different set of challenges. The focus should shift from proving the technology works to building the structures that allow it to scale safely and consistently. That means shared governance instead of departmental reinvention, funding models that reflect how AI capability grows, and accountability that can withstand real scrutiny.

Success shouldn’t be measured by the number of AI pilots an agency has run. It should be measured by whether AI is helping agencies deliver better outcomes for people — whether that’s faster claims processing, more consistent decision-making, or more time for frontline staff to focus on the parts of their role that genuinely require human judgement. Just as importantly, those gains must be achieved while maintaining trust, accountability and appropriate oversight.

Australia did not become a global leader in digital government by chasing novelty. It did so through consistent execution over time. AI now needs the same approach.

Image credit: iStock.com/Sandwish