Government's future workforce challenge is not simply an AI challenge

When we talk about the future workforce, it is easy to make AI the headline, but for government, the real leadership challenge is not choosing the technology, but helping people understand how their work will change, where accountability sits, and how teams can use new capability to deliver better outcomes for citizens. This challenge starts with clarity: on the outcomes we are pursuing, the decisions that need to be made, who owns them and how teams work together to deliver them.

AI adoption is widespread, yet meaningful application of the technology remains limited. McKinsey’s ongoing ‘State of AI’ research shows 78% of organisations now use AI in at least one business function, yet only 21% have redesigned even part of a workflow to take advantage of it. Adoption is no longer the hard part. What matters most is what happens after the pilot: who owns the decision that follows, how the work actually changes and whether any of it improves outcomes for the people relying on the service.

For government agencies already managing workforce constraints, rising service expectations, legacy systems and competing policy priorities, this is no small task. AI can help, but technology alone will not build a more capable, future-ready public sector workforce. The organisations most likely to realise value are those that establish clarity through three connected leadership shifts: moving from experimentation to impact, starting with the work rather than the tool, and connecting investment to delivery.

1. Move from experimentation to impact

Most agencies are already testing AI somewhere. That is a reasonable first step, but a pilot is never the outcome. It is an experiment that shows whether a technology deserves investment and scale or should be discarded.

Experimentation should begin with a clear hypothesis: what are we trying to improve, for whom, and how will we know it worked? The answer might be faster response times, better access to trusted information for frontline staff, less administrative work or a better citizen outcome. That benefit should be defined up-front, not reverse-engineered afterwards to justify the spend. Start by agreeing a baseline and the intended improvement, whether reduced cycle time, fewer errors or lower cost, before choosing a solution.

This means creating room for teams to test and learn, while holding clear boundaries around data, risk and human oversight, particularly where decisions touch people’s health, safety or access to services. Planview’s own 2025 analysis reinforces this: high-performing organisations that balance speed, adaptability and governance were, on average, 65% more likely than laggards to report improvements across nine performance areas, including risk reduction and time to market. Move too cautiously and innovation stalls. Move too fast and you lose visibility and trust. The job is to create enough freedom to experiment and enough discipline to turn what is learned into measurable impact.

2. Start with the work, not the tool

The right question to kick-start experimentation is not ‘where can we use AI’, but rather ‘which citizen, staff or operational experience needs to improve?’

That distinction matters more than it sounds. Too many organisations start with the technology and go looking for a significant problem to solve, producing a pile of tools and proofs of concept that change little for anyone. Instead, begin with the work: look at where low-value administrative work is consuming skilled people’s time, where fragmented information and manual hand-offs are creating delays, and where faster access to trusted information would improve a decision. Anchor this in strategic priorities and performance gaps, and look hardest at work that is slow, costly, inconsistent or difficult to scale — especially where decisions involve prioritisation, risk, resource allocation or forecasting. Those are the problems worth solving, with benefits that compound.

Then look at the end-to-end workflow rather than an isolated task. Map where information is gathered manually, where decisions are made without the right context and where hand-offs create delays, because improving one step may simply shift the bottleneck elsewhere.

AI can help summarise information, support staff with trusted knowledge, or automate low-risk administrative steps; but it should not replace accountability for decisions that need professional judgement, empathy or care — and in government that line matters more than most industries. The strongest opportunities embed AI in existing processes, systems and accountability, turning insight into action while preserving human judgement and controls, rather than operating as another standalone experiment.

McKinsey’s public sector research makes the same point from a different angle: it identifies workflow redesign, not the technology layered on top of it, as the biggest source of AI value. A process that has been inefficient for years does not suddenly become effective because AI has been added to it. Here is a useful litmus test for any potential use case: if you took the technology away, would everyone still agree the work needs redesigning? If not, it is probably not ready for an AI solution.

Once the problem is clear, prioritise use cases against value and feasibility. Weigh expected impact and alignment with strategic objectives against data quality, process readiness, risk and explainability, as well as the ability to scale across teams. Getting this right means bringing the people closest to the work into the design process, not just technology teams. Frontline staff know where the friction is. Service leaders know the intended outcome. Data, privacy and risk teams know what safe looks like. When these groups work from a shared outcome rather than a preferred platform, agencies are more likely to build services that staff can use with confidence and citizens can trust.

3. Connect investment to delivery

An AI strategy is only credible if leaders can see from the start what is funded, what it is meant to deliver, and whether it is working. Without that visibility, AI becomes a collection of costly, disconnected pilots, and no one can say with confidence what to scale, what to stop, or where the next dollar should go.

The gap between AI ambition and execution is already evident in Australian governance research. A 2025 survey by the Governance Institute of Australia found that 93% of respondents could not effectively measure the return on investment from AI, while 88% had struggled to integrate generative AI with legacy systems. The implication is clear: investment and experimentation alone do not create value.

The fix for this is treating AI and digital initiatives as a connected portfolio at the outset rather than a set of isolated projects. A shared view across strategic priorities, funding, capacity, delivery milestones and risk allows leaders to see which initiatives are advancing citizen outcomes, what should be paused to free up capacity, and what foundational investment in data, security, workforce capability and change management still needs to happen. Leaders need to govern that portfolio actively, funding the highest-value opportunities, setting clear criteria for moving from trial to scale, and stopping or redirecting work that does not demonstrate progress.

It also sharpens workforce choices, including where people need new skills, where specialist capability should be retained and where AI can safely take on repetitive work. Just as important, every use case needs an executive sponsor, a business owner, an accountable delivery team, defined decision rights and explicit responsibility for benefits realisation.

Measure more than adoption or activity. Track baseline performance, business impact, cost to operate, quality, risk and realised benefits, and link each investment to the outcome it was meant to produce.

Governance also needs to move at the speed decisions now require, rather than being stuck in slow, sequential committee cycles that were built for a slower kind of change. Controls should be proportionate, lightweight for low-risk applications and more rigorous for high-impact decisions. That means clear risk tiers, approval gates, human oversight, security and privacy requirements, auditability and, where needed, retirement procedures. It also depends on trusted, accessible data and reusable platforms, because disconnected tools rarely scale.

Leading people through the shift

The final discipline is making AI part of normal everyday work. That means redesigning roles, hand-offs, incentives, skills and management routines rather than simply adding a model to an existing process, then monitoring performance after launch and improving both the workflow and the AI capability as teams learn from use.

Collectively these connected shifts reinforce one another. Experimentation creates learning. Workflow-led design makes that learning relevant to the people it is meant to serve. Portfolio discipline turns it into delivery that can scale. What binds them together is leadership clarity on outcomes, ownership and the boundaries within which people and AI operate.

The future public sector workforce will not be defined by which tools it adopts. It will be defined by whether its leaders can bring people, investment and delivery together around outcomes that matter most.

Image credit: iStock.com/filadendron