When it comes to AI, public sector leaders are accountable to taxpayers twice. Once for how efficiently they run government, and again for what happens to the people who work in it, who are themselves taxpayers.
So when New Zealand's Finance Minister Nicola Willis announced that nearly 9,000 public service roles will be cut over three years, with AI adoption listed as a primary mechanism for delivering $2.4 billion in savings, it caused quite a stir. The reaction landed roughly where you'd expect. Unions alarmed, the opposition noting that more than half those jobs are outside Wellington, held by social workers, border staff, and people working in the conservation estate. Some experts applauded the efficiency logic.
Both sides are probably right, which is exactly the problem.
Move too slowly on AI and you're accused of mismanaging public money, running outdated systems while the private sector pulls ahead, leaving your own workforce unemployable in a labour market that's moving without them. Move too fast and you're accused of treating public servants as a line item, announcing savings before you've done the work, using AI as political cover for cuts that were always coming. There's no clean path through that, there's just the question of whether you navigate it with empathy and rigour or announce your way through it.
The NZ announcement, at this stage, looks more like the latter. The savings figure is precise, the detail is not. Which 9,000 roles? Which tasks within those roles will actually be absorbed by AI capability? Which will disappear because the work goes away? And which are being cut on the assumption that AI will fill the gap, before anyone has confirmed it can?
That last question matters. In 2023, New York City launched an AI chatbot to help small business owners navigate city regulations. The mayor called it a landmark deployment but when released, it told business owners they could legally take a cut of their workers' tips, in direct contradiction of labour law. The city kept it running for two years, adding disclaimers while defending the technology. In January 2026, the incoming mayor killed it, describing it as "unusable." Safe to assume the team it replaced are long gone.
That's not an argument against AI in government, but it is an argument for understanding what you're deploying, and what the impacts are, before you announce the outcomes.
Why AI workforce transitions are harder in the public sector than the private sector
The public sector has structural constraints that make this harder than private sector AI adoption, not easier. Procurement cycles that run months behind the technology, legacy systems that don't talk to each other let alone to new AI tooling. And then there's the employment law reality that tends to get left out of efficiency announcements entirely.
Public sector employees carry significantly stronger protections than their private sector counterparts, and many permanent employees have a constitutionally protected interest in not being terminated without due process. In practice, most are covered by enterprise agreements that include redeployment obligations, minimum consultation periods, and redundancy provisions that require genuine justification, not just a budget rationale. Every one of those 9,000 redundancies will need to be individually defensible. If the justification is "AI will absorb this work," someone will test that, and a pre-Budget speech won't be sufficient evidence.
The redeployment obligation is where this gets particularly important for reskilling. You can't credibly demonstrate that redeployment was genuinely explored if you haven't done the task-level work.
Which tasks in the affected roles are changing rather than disappearing? Where does the capability already exist in your workforce to move toward the work that remains? Which people, with some targeted development, could transition into the roles that AI creates rather than eliminates?
These aren't soft HR questions, they're the legal and operational groundwork for a transition that holds up. Announcing a number without that analysis doesn't just create a communications problem. It creates a process that's vulnerable at every individual redundancy decision.
And a workforce displaced without that groundwork doesn't just accept it and move on. They file for benefits, strain regional economies, and appear in the statistics that other parts of the public service then have to manage with fewer people.
What governments should do before cutting public sector jobs using AI
Task-level analysis of AI's impact on public sector roles should precede any workforce reduction announcement. This means mapping each role into its constituent tasks and assessing each task across five categories:
1. tasks that stay with humans because they require judgment, relationships, or ethical reasoning AI cannot replicate;
2. tasks where humans lead and AI assists;
3. tasks where AI leads and humans review;
4. tasks likely to be fully automated;
5. and tasks that become obsolete entirely.
That breakdown, applied at scale across an organisation's actual roles, is what produces a defensible workforce plan rather than a defensible savings figure.
GoFIGR's enterprise AI Impact Assessment does exactly this, using structured task extraction, AI capability assessment, and its custom skills taxonomy to map workforce exposure and skills adjacency across an entire organisation, without surveys, without announcements, and without involving employees until there's something concrete to say.
One AI expert put it plainly: leaders who are genuinely literate about this technology make better informed and strategic decisions about what it means for the people working for them, and ultimately, for the people they serve.
Literacy here doesn't mean knowing how to use Microsoft Copilot. It means knowing which tasks in your organisation are well suited to AI, which require human judgment by definition, and which you'd be cutting on hope rather than evidence.
Willis is right that in too many parts, the back-office of government still runs on old-fashioned systems with slow bureaucratic processes that are too often about box-ticking rather than improving outcomes. That's a fair description of most large public sector organisations and a genuine cost borne by citizens. The question isn't whether to change it. The question is whether a headcount reduction is the plan, or the outcome of one.
Those are different things.
A plan starts with task-level analysis: which tasks within which roles are being absorbed by AI, which are being restructured because the work is changing, and which people have adjacent capability that makes reskilling a better bet than redundancy.
An outcome is what you announce when the savings target comes first and the methodology is implied.
The workers caught in this are not abstractions or cost lines. They're the same people whose taxes fund the public services their former colleagues will now have to deliver with fewer hands. The stewardship argument cuts both ways, and any government serious about AI workforce transition eventually has to reckon with that.
The brief isn't impossible. It's just harder than it looks from a pre-Budget speech. Do the prep work first. The number, and the workforce plan, follow from that. Not the other way around.
If you're leading an AI transition in the public sector and need analysis that holds up legally and operationally, GoFIGR's enterprise AI Impact Assessment maps which tasks within your roles are affected by AI, which aren't, and where your workforce has the skills adjacency to move.
No surveys, no announcements, no disruption while you're still figuring out the plan. Talk to us at gofigr.ai or try it for free here.
Helena Turpin spent 20 years in talent and HR innovation where she solved people-related problems using data and technology. She left corporate life to create GoFIGR where she helps mid-sized organizations to develop and retain their people by connecting employee skills and aspirations to internal opportunities like projects, mentorship and learning.
