The NZ AI Benchmark / The framework
The framework
The five stages and the six steps.
Everyone is adopting AI. The winners are adapting. Adoption makes your existing business faster. Adaptation redesigns the business AI makes possible. You need both, and only adaptation changes your competitive position.
The NZ AI Benchmark measures where an organisation sits on that path using the five stages and six steps of the AI Playbook, published by AI New Zealand. Maturity is not measured by how many tools you have. It is measured by how much you have been willing to redesign the work itself.
Five stages
The five stages.
Each stage has a description, a trap, and a sensible place to start. The diagnostic question is deliberately unflattering: pick the row that sounds like a normal Tuesday, not the row you aspire to.
Stage 1
Aware
What it looks like
Leaders reading headlines and feeling pressure. A few have tried ChatGPT at home. Curiosity, but no permission, no plan, no shared understanding.
The trap
Mistaking awareness for progress. The cost is the quiet one: the gap widens every week you wait.
Where we start
Lunch & Learn, CEO briefing.
Stage 2
Experimenting
What it looks like
ChatGPT bookmarked, Copilot activated. A few people doing impressive things, most are not. No policy, no standard, no visibility on what works.
The trap
This is the highest risk stage, not the safest. Your best people are feeding company context into ungoverned tools, and the wins live in whoever is using AI that day.
Where we start
Executive AI Briefing, AI Discovery Workshop.
Stage 3
Adopting
What it looks like
Tools rolled out properly. A policy, an ambassador, training, and solid use cases delivering real time savings. Feels like success.
The trap
The adoption ceiling. You made the legacy way faster without asking whether the legacy way is still right.
Where we start
Use case scoping, AI Sprint, workflow mapping.
Stage 4
Adapting
What it looks like
You stop asking which tool and start asking how the business should sense, think, decide and act with AI in the loop. Methodology, brand voice and workflows encoded once, connected to real systems, used by everyone, improved monthly.
The trap
Trying to adapt everything at once. Adaptation is operating model surgery. Choose the few workflows where the prize justifies the redesign, and adopt everywhere else.
Where we start
AI Roadmap, AI Operating System.
Stage 5
Frontier
What it looks like
Populations of agents running adapted workflows. New roles appear: people who own agent quality, set risk appetite, and assure agents in production the way a factory assures its line.
The trap
Forgetting the discipline. Frontier is not a licence to skip governance. Agent assurance is a new budget line, not an optional extra.
Where we start
Retainer, agent assurance advisory.
Where most people actually sit: between Experimenting and Adopting, believing they are further along than they are. The next move is always one stage up, not five.
The gap
The Adaptation Gap.
The gap is the line between Stage 3 and Stage 4. Between making the old work faster, and redesigning the work itself.
87 percent of New Zealand organisations use some form of AI and only 12 percent have scaled it across the whole business (Datacom State of AI Index 2025).
That distance is the Adaptation Gap, and it is where almost every New Zealand organisation is currently stuck. AI does not fix a broken process. It accelerates it. So the strategy question was never which tool. It is how you want the business to run when AI is in the loop.
That is not an adoption problem, it is an operating model problem.
Governance sits on the commercial side of this ledger too. 62 percent of New Zealanders say they would stop using a product or service over concerns about how the organisation uses AI (One NZ AI Trust Report, May 2026). Clear rules are what let a manager say yes to putting AI in a real workflow, which is why organisations with them tend to move faster, not slower.
Six steps
The six steps, at two altitudes.
Six moves take a business from Aware to Adopting. The same six, done deeper, carry it into Adapting and the Frontier. The most common mistake is treating them as a checklist you finish once. They are a discipline you keep levelling up. The benchmark scores how far up each one you are.
Assign an AI Ambassador.
Adoption altitude, to Stage 3
One person leads education and adoption, gets tools into hands safely, gives nervous staff their first win. Choose the person who knows your customers and your process.
Adaptation altitude, Stages 4 to 5
Becomes an AI council and a centre of enablement. Leadership drives it together, aligned to the business plan. Not delegated: federating it out to run independently does not scale.
Set governance protocols.
Adoption altitude, to Stage 3
A clear, organisation wide AI policy. A statement of what you will and will not do with AI, so staff can make confident, responsible decisions.
Adaptation altitude, Stages 4 to 5
Grows into agent assurance. Once agents act in production and talk to other agents, you need a quality system: sampling, audit, and assurance profiles that tighten when an agent drifts.
Identify practical use cases.
Adoption altitude, to Stage 3
Workshops to find quick wins. A few repetitive, time consuming tasks to build capability, get value on the board fast, and build belief.
Adaptation altitude, Stages 4 to 5
Becomes workflow redesign. Map how a critical workflow senses, thinks, decides and acts, then rebuild it around AI rather than bolting AI on. Move a metric, not a use case.
Understand your risks.
Adoption altitude, to Stage 3
Stand up an AI committee. Review data, IP, security and privacy against how employees could actually use AI tools.
Adaptation altitude, Stages 4 to 5
Becomes a leadership discipline: an explicit risk appetite for how close to the frontier you will operate, and owned assurance so problems are caught early. Prompts are not controls. Real controls are technical.
Select the right AI tools.
Adoption altitude, to Stage 3
A selection framework prioritising security, scalability and integration with your data. Favour tools that serve multiple use cases. Avoid the shiny object trap.
Adaptation altitude, Stages 4 to 5
Tools become a system: a shared AI operating system with governance, process, knowledge, integration and skills layers. For some, sovereign AI matters here: models curated for your culture, values and customers.
Provide ongoing AI skills training.
Adoption altitude, to Stage 3
Practical, hands on training so the team can use the tools and operationalise new workflows. Power up fast adopters with advanced skills.
Adaptation altitude, Stages 4 to 5
You build an AI native workforce. Roles deepen, they do not disappear. Frontline staff own agent quality, experts move from firefighting to calibrating, leaders invest in literacy across the whole organisation.
Method
How the benchmark scores.
Forty questions, five per dimension, across eight dimensions of how the business runs. Each question offers five descriptions and you pick the one closest to today. Weighted scores roll up into a dimension score and an overall stage. Scoring is rules based and runs on our servers, so the same answers always produce the same result.
Not sure is a valid answer to every question. It is left out of the dimension score and counted as a visibility gap instead, so honesty is never penalised. A dimension with fewer than two answered questions is reported as not assessed rather than scored.
Benchmarking happens in two layers. The first compares you with the five stages and best practice, and it is available immediately. The second compares you with other New Zealand organisations of a similar industry, size or region. That layer switches on for a segment once at least 20 organisations in it have completed the benchmark. Until then we say the comparison is not ready rather than show a number built on too few responses. Respondents to the benchmark are not a representative sample of New Zealand business, and every peer view says so.
The New Zealand questions draw on the Privacy Act 2020 and Privacy Commissioner guidance, MBIE's Responsible AI Guidance for Businesses, the Algorithm Charter for Aotearoa New Zealand, Te Mana Raraunga principles on Māori data sovereignty, and the AI Forum's AI Blueprint for Aotearoa.
Credits
Who built what.
The five stage model, the Adaptation Gap and the six steps are AI New Zealand's published AI Playbook. That is the thinking the benchmark measures against.
The scoring engine and the question architecture, including the eight dimensions and the weighted five option scale, were originally developed by Cairn and are adapted here with permission. We have recalibrated the stage names, thresholds, commentary and recommendations to the AI Playbook. Credit where it is due: the mechanics are theirs, the position is ours.
Free, 20 minutes
How does your business stack up?
Forty questions, your stage and your next move on screen straight away.