Trust layer

Methodology

The score is arithmetic. No model decides your band, nothing is inferred about you, and the same answers always produce the same number. This page is the whole method — if you disagree with a cut-off, you can compute your own.

How the number is built

12questions, four options each

0–3points per question

36maximum score

Each question is answered by picking one of four statements: not in place (0), one person or one try (1), used but not documented (2), part of the process (3). The four statements exist so the score measures whether something is in the process, not whether it exists in someone's head. Missing answers count as zero, and the report shows you every answer with the score it got.

What each dimension covers

DimensionQuestionsWhy it is weighted this way
Data & privacy
Tool ledger & control
Work processes
People & policy
Measurement & review

Bands

ScoreBandWhat it means
0–9 Shadow AI It is happening whether or not you know about it.
Ad hoc Someone is experimenting. Nobody is accountable.
Defined You are running AI, but you cannot prove it.
36–36 Institutionalised AI is a process with a name attached to it.

The hour estimate

The report converts your point deficit into an indicative monthly figure of 1.5 hours per missing point, then multiplies that by NZD 74.00, the building small-firm output rate for 2026 Q1. Both halves are rough: the hours factor is a planning figure, and the hourly rate moves with the quarter. It is labelled an estimate on every report it appears on, and it is never presented as a saving you should expect.

Privacy of the score itself

The score runs in your browser. Nothing is transmitted unless you type an email and press the store button, at which point your answers and that email are written to a Cloudflare KV store under the site's privacy policy, for the purpose of sending you the report. We do not run your answers through a language model, we do not use them for training, and we do not sell them.

Sources

SourceUsed for
embedAI — State of AI in New Zealand Business (2026-04)The 87% → 12% adoption gap.
Microsoft / EMA workforce survey (2026)81% staff bring their own tools; 13% of SMEs have a policy.
Deloitte Access Economics for 2degrees (FY25)NZD 400,000 average earnings difference for AI-adopting SMEs.
Xero small business productivity benchmarks (2026 Q1)NZD 74.00 per hour for building small firms.
Kennards Hire — Construction Confidence Check 202659% worried about closing; 35% crediting AI and digital.
NSP — Shadow AI in Construction Firms in New ZealandThe documented behaviour this whole site exists to fix.
AI Forum New Zealand — AEC working group (2026-05)Industry position on formalising AI policy.
MBIE / NZ Government — AI strategy (2025-07-08) and responsible AI guidanceOfficial positioning used in the PP12 path.
business.govt.nz — Data for businessBusiness counts by industry; the 81,000 construction figure.
Vendor pricing pages (per tool)Indicative NZD costs and data regions in the ledger.

Change control

Every figure on this site lives in the repository with a date. Prices in the ledger are marked with the month they were last verified. When a source is superseded, the new number arrives with both the old one and the date change noted. Anything a model drafts goes into a pull request and waits for a human.

Corrections. Found something wrong? hello@nzai.online — we fix it, we date the fix, and we say so in the next email.