New Zealand Government AI: $200 Million in Strategy, 55 Operational Use Cases to Show for It
The data tells a clear story. Despite frameworks, funding and political will, New Zealand’s public sector AI adoption is stuck in a cycle of pilots, caution and quietly used personal accounts.
The public sector AI story sounds impressive until you look underneath
There is a version of the New Zealand government AI story that sounds genuinely impressive.
Seventy agencies surveyed. A national AI strategy launched. A Public Service AI Framework in place. Significant investment in tuition, training and advanced technology. Masterclasses for leaders. Foundational courses for public servants.
Read the press releases and you could think the New Zealand public sector was well on its way to becoming an AI-powered operation.
Then you look at what is actually happening on the ground.
According to the 2025 cross-agency survey for artificial intelligence use cases, 70 government organisations reported 272 AI use cases. Only 55 were deployed and in operational use.
That means most identified AI projects are still sitting in planning, trialling or ideas phases. The Government Chief Digital Officer’s public AI material also shows the scale of AI guidance now available to agencies, including the Digital.govt.nz artificial intelligence guidance hub.
The use cases that have made it to production are overwhelmingly low-risk: summarising feedback, automating meeting notes and basic search. Those are useful, but they are not core service delivery. They are not decision support. They are not the kind of work that transforms how government operates.
This is not a technology problem. It is an institutional one. And until someone names it clearly, the cycle of expensive strategy followed by cautious inaction will continue.
Key point: New Zealand’s government does not have an AI strategy problem. It has an AI execution problem.
The numbers behind the narrative
The Government Chief Digital Officer’s 2025 cross-agency survey is one of the clearest public pictures of AI activity in New Zealand government.
The headline numbers look like progress: 272 use cases reported by 70 government organisations, up from 108 use cases across 37 organisations the previous year.
But the detail underneath tells a different story.
Fifty-five operational use cases across 70 agencies means most agencies have zero or one AI project actually running in production.
The majority of reported activity sits in categories like idea, trial or planning. These categories generate survey responses, but they do not necessarily generate outcomes.
The operational use cases cluster around safe, low-stakes applications. Document summarisation. Meeting transcription. Internal search.
These are useful, but they are the AI equivalent of using a smartphone only for phone calls.
The capability exists to do vastly more, and the gap between what is possible and what is permitted is enormous.
Compare this with the private sector. Datacom’s 2025 State of AI Index reported that 87% of New Zealand organisations surveyed are now using some form of AI in their operations.
The gap is not because the technology does not exist. It is because public sector adoption is moving through a much narrower institutional pathway.
Useful distinction: Reported AI activity is not the same as operational AI adoption. A use case in planning does not improve a service, reduce a bottleneck or change how work gets done.
The Defence case study: policy as paralysis
The Ministry of Defence and New Zealand Defence Force illustrate the problem at its most extreme.
As of mid-2025, the NZDF confirmed it was not using AI operationally or in production. Staff were prohibited from using AI-generated information for work without explicit approval from a Deputy Secretary. Their only autonomous system, the Phalanx close-in weapons system, predates modern generative AI by decades.
There are legitimate reasons for caution in a defence context. Security classification, operational sensitivity and the consequences of AI errors in military environments are real and serious.
But a blanket prohibition approach creates its own risks.
It means the organisation builds almost no internal capability. It means that when AI becomes unavoidable, the institutional knowledge to deploy it safely will not exist. It also means the gap between New Zealand’s defence capability and allies that are actively integrating AI widens over time.
Useful distinction: Caution is not the same as governance. Good AI governance enables safe use. It does not simply defer the problem.
The Defence approach is not strategy. It is deferral dressed up as caution.
The international context: last to move, slow to follow
New Zealand released its first national AI strategy in July 2025.
MBIE describes New Zealand’s AI Strategy: Investing with confidence as adoption-focused, with an emphasis on applying proven AI solutions rather than building foundational AI models domestically.
That is a sensible strategic choice for a small economy. New Zealand does not need to compete with frontier model labs to benefit from AI. It can adopt, adapt and apply proven technology intelligently.
But adopting implies actually adopting.
The strategy talks about creating an environment where businesses can invest, encouraging adoption and removing barriers. These are the verbs of facilitation rather than action. They describe a government that wants to be seen as supportive without necessarily taking on the institutional risk of going first.
Oxford Insights’ Government AI Readiness Index provides a useful external benchmark for comparing how governments are positioned for AI adoption and public-sector capability.
Whether New Zealand is viewed through strategy documents, readiness benchmarks or agency-level use cases, the same pattern appears: the policy environment is improving, but implementation is still constrained.
The skills gap is real, but it is not the whole story
Internal and market data both point to capability challenges.
MBIE’s material on AI uptake references skills and expertise as major barriers. Datacom’s AI research has also identified lack of expertise and internal capability as adoption barriers for New Zealand organisations.
These are real problems.
But the skills gap is also being used as a convenient explanation for something more fundamental: a cultural unwillingness to accept the risk that comes with doing something new.
You do not need deep AI expertise to use a summarisation tool. You do not need a data science degree to draft correspondence with AI assistance. The skills required for the low-risk, high-value applications that should be basic across government are modest.
What is missing is not just capability. It is permission.
The Deputy Secretary approval requirement at Defence is an extreme example, but the same pattern appears in softer forms across the public sector. Risk-averse cultures create approval bottlenecks. Approval bottlenecks create delays. Delays become the status quo. The status quo then gets reframed as responsible adoption.
Practical point: Capability matters, but permission structures matter too. If the safest pathway is always “do nothing until approval is granted,” AI adoption will remain slow by design.
Shadow AI: the elephant in the room
While official channels move cautiously, public servants and knowledge workers are making their own decisions.
Across organisations, workers are already using generative AI tools to summarise documents, draft correspondence, brainstorm ideas, rewrite emails and speed up routine work. Some of that use is sanctioned. Some of it is not.
This is shadow AI: unsanctioned, ungoverned and invisible to the people responsible for data security, quality assurance and policy compliance.
The risk is obvious. Staff may upload internal documents to public tools, paste sensitive information into free-tier AI services, or use outputs without any review, audit trail or source checking.
The irony is sharp. The same organisations that will not approve AI pilots because of data security concerns can create conditions where data leakage happens anyway, completely outside any governance framework.
Shadow AI is not simply a failure of staff discipline. It is a failure of institutional leadership.
When the official process for using AI is effectively “do not use it unless a senior executive gives permission,” people who need to get work done will often find another way. The productivity advantage is too large to ignore, and the tools are too accessible to block completely.
The answer is not tighter restriction. The answer is sanctioned, governed access to AI tools with clear guardrails, so that usage already happening in the shadows moves into a framework where it can be monitored, improved and made safe.
Shadow AI grows in the gap between what workers need to do and what institutions permit them to do safely.
The “sinking lid” trap
There is a revealing pattern in how some agencies are thinking about AI’s role.
Rather than using AI to expand capability or improve services, many are focused on what has been described as a sinking lid approach: using AI to upskill remaining staff as positions go unfilled, rather than replacing departing workers.
On paper, this sounds pragmatic. In practice, it is a recipe for stagnation.
It positions AI as a cost-management tool rather than a capability multiplier. It means the ambition is to maintain current service levels with fewer people, not deliver better outcomes. It also frames AI defensively, protecting what exists rather than asking what could be improved.
This is the institutional equivalent of using AI to tread water instead of learning to swim.
Government agencies facing budget pressure and staffing constraints have a legitimate need to do more with less. But if AI is framed only as a way to absorb pressure, its ceiling is capped before anyone has tested what it can actually improve.
What would actual adoption look like?
The gap between where New Zealand government sits and where it could be is not a technology gap.
The tools exist. The frameworks exist. The funding exists. What is missing is the institutional willingness to move from strategy to execution at pace.
Actual adoption would mean every agency having at least one AI-assisted process in core service delivery, not just meeting notes.
It could include:
It would mean procurement frameworks that allow agencies to trial AI tools without a twelve-month approval process.
It would mean measuring AI adoption by outcomes delivered, not use cases reported.
It would also mean accepting that some implementations will fail, some outputs will be wrong and some pilots will need to stop. The cost of learning from those failures is much lower than the cost of not starting.
Critically, it would mean addressing shadow AI directly. Not by banning it harder, but by giving public servants access to secure, sanctioned AI environments with clear rules about what data can and cannot be used.
Practical next step: Agencies need to move from passive frameworks to active AI strategy, implementation roadmaps, governed tool access and targeted workflow automation pilots.
The consulting opportunity, and the honest assessment
For organisations like Changeable, this gap between strategy and execution is exactly where the work needs to happen.
Not more frameworks. Not more strategies. Practical, on-the-ground support that helps agencies move from their 55th pilot to their first scaled deployment.
That means governance that enables rather than blocks. Change management that builds confidence rather than compliance. AI literacy programmes that start with the actual work people do, not abstract concepts about responsible innovation.
It also means being honest about what we are seeing: a public sector that has invested heavily in the appearance of AI readiness while remaining, in practice, cautious to the point of paralysis.
The strategy is there. The funding is there. What is missing is the courage to do it, and the practical support to do it well.
New Zealand’s government does not have an AI strategy problem. It has an AI execution problem. Every month that the majority of identified projects remain in planning is a month where the gap between what is possible and what is permitted gets harder to close.
What Changeable helps with
Changeable helps organisations move from AI ambition to practical, governed implementation.
Start with a Decision Clarity Session
A Decision Clarity Session is a no-obligation conversation where we listen to what you are trying to achieve, what is getting in the way and whether AI, automation, process improvement or governance is the right next step.
Frequently asked questions
Why are there so few operational AI use cases in New Zealand government?
The public material suggests a mix of caution, capability gaps, approval friction and risk sensitivity. Frameworks and strategies exist, but many projects remain in planning, trial or idea stages rather than production.
Is caution around government AI a bad thing?
No. Caution is necessary where public trust, privacy, security and decision-making are involved. The problem is when caution becomes paralysis. Good AI governance should create safe pathways for use, not stop all meaningful progress.
What is shadow AI?
Shadow AI is the use of AI tools without formal approval, oversight or governance. It often happens when staff see productivity benefits but the organisation has not provided safe, sanctioned tools or clear rules.
What should agencies do first?
Start with low-risk, high-value workflows where AI can reduce friction without making final decisions. Then build governance, data handling rules, human review points and measurement into the process from the beginning.
Does New Zealand government need more AI strategy?
The central issue is not more strategy. The bigger problem is execution: turning frameworks, guidance and pilots into governed operational use cases that improve real work and services.
How can AI governance enable adoption?
Good governance creates clear rules for approved tools, data handling, human review, risk escalation, quality checking and accountability. It should help people use AI safely rather than simply making use difficult.
Can Changeable help public sector organisations with AI adoption?
Yes. Changeable helps organisations clarify use cases, build AI strategy, design governance models, improve workflows and move from pilots to practical implementation.
Move from AI frameworks to governed implementation.
Changeable helps organisations turn AI ambition into practical adoption through clear use cases, enabling governance, workflow automation, data models, AI agents and implementation support that works in the real operating environment.