Skip to main content

An AI for Good New Zealand initiative

AI Policy Sandbox

A transparent NZ sector-calibrated framework for testing AI policy tradeoffs under uncertainty.

The problem

New Zealand's AI policy conversation rests on a fragmented evidence base. Adoption figures in circulation range from 32% to 87%. No two sources measure the same thing. Some widely-cited figures are not adoption measures at all.

Policy built on any single number is standing on weak ground. And when that number is applied uniformly across an economy as structurally diverse as New Zealand's - from agriculture to financial services to public health - the mismatch between policy design and sector reality gets worse, not better.

There is currently no tool that lets anyone test: what happens if we invest here vs there?

What the sandbox does

AIPS replaces single-number thinking with a structured, sector-calibrated framework that compares different policy designs under uncertainty.

It covers all 19 sectors of the New Zealand economy, classified using the official ANZSIC standard (Australian and New Zealand Standard Industrial Classification) - the same system used in official GDP and employment statistics. The structure is tiered: 9 full explanatory sectors, 6 simplified sectors, and 4 residual sectors. The whole economy is represented, because honest aggregate-policy comparison demands a whole-economy denominator.

Scenario A

Aggregate policy

Broad economy-wide allocation by GDP share. The default approach most policy conversations assume.

Scenario B

Targeted demand-side

Support focused on sectors where adoption is lagging, bottlenecks are acute, or direct intervention unlocks productivity.

Scenario C

Targeted supply-side

Investment in enabling capacity - technology, skills, infrastructure, and diffusion mechanisms that raise adoption across the economy.

Sector structure

Each sector represents a distinct AI adoption archetype. The differences are not cosmetic - they change how policy works.

Sector Archetype Why it matters
AgricultureTargeted-policy beneficiaryNZ's most distinctive sector. No published adoption rate. Long diffusion timeline.
ManufacturingProductivity sweet spot58% adoption. Clearest productivity case.
Professional ServicesDiminishing returns testHighest adoption. Further investment yields smaller marginal gains.
Public SectorEmployment storyLowest adoption, fastest acceleration. Capability, culture, and scale are the barriers.
TechnologySupply-side enablerNot just an adopter - it enables the other sectors. $24B GDP, $11.4B exports.
HealthcareEquity-constrained62% adoption. Admin AI vs clinical AI. Equity and governance are binding constraints.
ConstructionWorst case for aggregateNo published adoption rate. 95% of firms under 10 people. Lowest digital maturity.
Financial ServicesRegulatory throttleHigh adoption constrained by dual regulators (FMA + RBNZ).
Retail & WholesaleConsumer-facing displacementLargest sector by weight. Wholesale 64% adoption. Retail = displacement story.

Plus 6 Tier 2 sectors (Education, Transport, Accommodation, Admin Services, Info Media, Utilities) and 4 Tier 3 sectors (Mining, Real Estate, Arts, Other Services) for whole-economy coverage.

Evidence base

The project draws on 10 completed research analyses and 38 catalogued data sources, including:

  • OECD "Miracle or Myth?" (2024) - productivity calibration
  • NZ Treasury AN 24/06 - structural barriers and diffusion lag
  • AI Forum NZ (Sep 2024) - adoption data and methodology critique
  • Datacom State of AI (2024) - 66% adoption (100+ employee firms)
  • PM's Chief Science Advisor (2023) - healthcare AI constraints
  • RBNZ "Rise of the Machines" (2025) - financial stability risks
  • NZTech Manifesto (2026) - tech sector baseline
  • Stats NZ (multiple) - GDP, employment, business demography

Current status

The project is in foundation phase. The analytical framework is built. The model equations are next.

Project scope locked
Whole-economy tiered structure defined (19 sectors)
Formal paper outline written
Methods note drafted
10 research analyses completed
38 data sources catalogued
348-cell parameter table created (35% populated)
Architecture decision made (ODE system model)
State-variable specification drafted
First-pass equations
Version 1 model build
Paper draft
External review
Public-facing interactive layer

Contributor FAQ

Common questions from potential contributors and collaborators.

What exactly is this project?

A transparent NZ sector-calibrated policy sandbox for testing AI policy tradeoffs under uncertainty. It compares aggregate versus targeted policy approaches across New Zealand's full economy.

Is this a forecasting engine?

No. It is a scenario model. It compares policy structures under uncertainty. It should not be presented as a machine for precise GDP or employment forecasts.

What kind of expertise are you looking for?

Sector specialists (agriculture, healthcare, construction, finance, retail, manufacturing, tech, public sector, education) who understand AI adoption dynamics. Economists with experience in system dynamics or productivity analysis. Policy analysts with NZ government or MBIE/Treasury experience. AI practitioners working in NZ industry who can validate adoption assumptions.

What is the time commitment?

Flexible. Advisory (2-3 hours for a structured interview), data contribution (share or point us to sector-specific data), model review (review assumptions for your sector), or co-authorship (deeper involvement in the paper).

Is this academic or commercial?

Academic. The target is a peer-reviewed publication and an open-access interactive web tool. No commercial interest.

Who is leading this?

The AI for Good New Zealand collective. This is a community-led initiative bringing together practitioners, economists, sector specialists, and policy analysts to build a tool that serves the whole of Aotearoa.

How do I get involved?

Email hello@aiforgood.org.nz with your name, which sector(s) you have expertise in, and what type of contribution you could make. Or explore the full project on GitHub.

Explore the project

The full scope, methods, data catalogue, and contribution brief are on GitHub. MIT licence.