How to choose an AI consultant in New Zealand: the show-me list
Ask a New Zealand AI consultant to produce four things before you pay: a client you can phone, a live system, a named builder, written exit terms.
Angus McDonald · 20 Aug 2026 · 8 min read
written for New Zealand ● also for Australia, United States, United Kingdom, Canada
Ask an AI consultant to produce four things before you pay: a client you can phone by name, a system in production you can watch a non-employee use, the name of the person who will do your work, and a written answer to what you keep if you stop paying. Answers about methodology are free. Only the four artefacts are checkable, and in New Zealand you can verify the business behind them on the free Companies Register in seconds.
Why most AI consultant vetting checklists test the wrong thing
Most AI consultant vetting checklists test what the consultant says. Ask about methodology, frameworks or certifications and you get answers, and answers are free. A market that buys on answers rewards fluency, not the ability to build a system that survives a staff change.
The questions that survive a bad supplier are the ones whose answers existed before you asked. Nobody can compose them in a meeting. That is the rule behind the Azgard Show-Me List: ask for the thing, not the answer about the thing.
The Azgard Show-Me List: four things to ask an AI consultant to produce
The Azgard Show-Me List is four artefacts you ask an AI consultant to produce before you pay. It is a list, not a score: three out of four is a conversation, one out of four is a decision.
| Ask them to produce | Why it works |
|---|---|
| A named client you can contact | It existed before you asked, and one call tests it |
| A system in production run by a non-employee | You are watching a habit, not a demo |
| The name and calendar of the person doing your work | Names commit, roles do not |
| A written answer to what you keep when you stop paying | A document exists or it does not |
Anything they can only produce after you pay is a promise, not evidence.
Ask for a named client you can contact, not a logo wall
The first item on the Azgard Show-Me List is a named client you can contact, and contact means a phone call, not a name on a website. A reference is a person who picks up and answers a question nobody scripted.
One objection is fair: some consultancies cannot publish client names without written permission. That is why the ask is a private introduction rather than a public logo. A firm bound by confidentiality can still ring a client and ask; a firm that will do neither has answered you. On the call: what broke, who fixed it, is it still running, and did the seller do the work.
Watch someone who is not the consultant use the system in production
The second item on the Azgard Show-Me List is a system in production operated by someone who does not work for the consultant. Two rival AI consultancies put it best: Kernel Flow's rule is that the key is understanding whether a firm builds running systems or just writes roadmap plans, and Team 400 tells buyers to ask how many AI systems they have running in production right now. Both rules predate the Azgard Show-Me List, and both firms pass their own test.
Do not accept the count, watch the thing. Ask for a screen share where a staff member at the client does a real task on an input nobody prepared. You are looking for boredom: a system people use is boring to them, and a demo is not.
Ask what you keep the day you stop paying
The fourth item on the Azgard Show-Me List is a written answer to what you keep the day you stop paying. Written, because you only test it once the relationship is over. Before you sign: whose name is on the model and software accounts, where the data sits, whether the automations keep running when the retainer stops, and whether somebody else could change the system without the consultant. You cannot buy a company brain - you can only be handed one that sat in your own accounts all along.
Measurement belongs to this item too. A separate artefact, the Azgard Three-Question Vendor Filter, asks what will be measured, against what baseline, and who reports it. Without a baseline taken before the build, you keep a system whose value nobody can defend.
How to verify an AI consultant's business is real in New Zealand
New Zealand gives a buyer more free, instant disclosure about a consultant's business than most vetting checklists ask for. None of these five searches needs a login.
| What you check | Where | What it discloses |
|---|---|---|
| The company | Companies Register company search | Status (registered, in liquidation, in receivership, removed), incorporation date, directors with names, addresses and appointment dates, shareholdings, and the full filing history |
| The person | Directors and shareholders search | Every company that person directs or holds shares in, past and present |
| Whether they are banned | Banned director search | People prohibited from managing a company, including under Australian law |
| Whether they are bankrupt | Insolvency Register | Bankruptcies, No Asset Procedures and Official Assignee liquidations, plus discharges in the last four years |
| Someone who is not a company | NZBN Register | Sole traders, partnerships, trusts and charities |
Two honest limits. The banned-director list is not exhaustive, and current bankrupts appear on the Insolvency Register. And the register publishes a director's residential address, which must be a physical street address rather than a PO Box, though the Companies (Address Information) Amendment Act 2025 will let directors substitute an alternative address, is not yet in force, and must commence by 18 November 2026.
What none of it proves is competence. New Zealand requires no licence or registration to work as an AI consultant, and management and IT consulting are absent from its list of occupations requiring registration by law. The register proves the business exists and who runs it, not what it can build.
The third-party vetting signal New Zealand buyers assume exists is the AI Advisory Pilot, and it is not one. The pilot programme is open and was expanded in May 2026 to a total of up to 150 businesses, up from 50, and extended to 31 January 2027; the funding arithmetic is on what an AI consultant costs in New Zealand. No approved-provider list exists for a buyer to read. Providers self-register with the Regional Business Partner Network against three criteria - located in New Zealand, professional indemnity cover of NZ$1 million or more, and an NZBN - and the network states there is no formal assessment process or KPIs you need to meet. A growth advisor shortlists providers and the platform sits behind a login, so advice to check a firm is MBIE-approved asks you to verify what you cannot see.
Borrow those criteria: ask any consultant where they are located, for a certificate of professional indemnity cover, and for an NZBN. The AI Forum of New Zealand says of its own AI Capability Directory that inclusion in the map does not imply an endorsement by us.
AI consultant red flags, and the ones that are not red flags
AI consultant red flags are mostly refusals: the seller will not name who does the work, the consultant drives every demo, nobody writes the exit terms down, and a price appears only after a paid discovery phase.
The ones that are not red flags: no AI-specific certification, because no such licence exists here; one person, provided that person is named; declining to publish client names while offering an introduction privately; and no New Zealand office when the work is remote.
Recourse here is cheap, and one clause removes much of it. The Fair Trading Act bans misleading and deceptive conduct in trade, but section 5D lets two parties both in trade agree in writing to contract out of it, usually as a non-reliance clause. So add a fifth ask: send me your standard terms, and show me the non-reliance clause. The unfair contract terms regime covers small trade contracts worth up to NZ$250,000 in any 12-month period, but only the Commerce Commission can act on an unfair term, not you. The Disputes Tribunal hears claims up to NZ$60,000, with filing fees of NZ$248 up to NZ$30,000 and NZ$496 above that, and no lawyer argues your case.
How Azgard answers its own four questions
Azgard is one engineer in Sydney. These are its four answers, including the ones that cost something. Item one: three named clients - Apex Signage, Home Grown Electrical and Inner Game Basketball - all Australian, and no New Zealand client. Item two: Apex Signage's quoting runs in production, operated by their own staff, with quoting time down from around three and a half hours to just over two per quote, a figure that fails fields four and five of Azgard's own Baseline Card: nobody recorded who took the timings, or on what date. Item three: the person doing your work is Angus McDonald, every time - a capacity ceiling, not a feature. Item four: every build is owned and documented so it runs without us, with no lock-in beyond an initial three-month term.
Two disclosures before you enquire. Azgard is Sydney-based and works remotely, with no New Zealand company, no local office and nobody on the ground; an Auckland working day and a Sydney one are two hours apart, which does not make Azgard local. Regional Business Partner Network providers must be physically located in New Zealand, so Azgard cannot register as one and Azgard's work cannot attract AI Advisory Pilot co-funding. Engage a New Zealand-based provider if that funding matters.
FAQ
tags: vendor-evaluation ● hiring ● due-diligence ● procurement
Angus McDonald
Founder, Azgard
Builds and operates production AI systems for organisations that need results, not slide decks.