How do you choose an AI consultant in New Zealand?
Short answer: start with the outcome you want in the next 90 days, then pick the firm type built for that job. A large IT firm when AI sits inside a wider technology programme. A boutique specialist when you want a working system in the tools you already run. A freelancer when the job is small and you can brief it. Then check three things: work they have already shipped, a named person doing the delivery, and a handover your team can run.
A report, a team workshop, a tool connected to Xero, and an engineer in the business for part of the week are different jobs. They suit different firms. The checklist, questions and red flags below apply whichever type you shortlist. If you are still choosing the tool, read which AI tool fits a NZ business first. A consultant should defend that choice against your stack.
What should you look for in an AI consultant?
Look for shipped work you can inspect, a named person who will do the delivery, and a handover your own team can run. A slide about transformation is not a substitute for any of the three.
- Finished work. A case study, a demo, or a system a client still uses. Ask to see it before you sign.
- A named person. Who will do the work, not only who will sell it. You want the name before the contract.
- Your tools. Xero, Microsoft 365, Google Workspace, the job system you already pay for. The value is the connection, not a new login.
- A line the AI does not cross. What it may draft, and what a person must approve before it sends, pays or files.
- New Zealand rules. The Privacy Act 2020 still applies. Ask where data sits, whether the vendor trains on it, and how GST is handled on software you will pay for.
- A definition of done. One deliverable, a date, and who signs it off.
- A way out. Documentation, logins in your name, and someone on your team who has used it.
What questions should you ask an AI consultant?
Ask who does the work, what they have shipped, what done means, and what happens to your data. If the answers stay vague, do not sign.
- Who will do this work, and what have they shipped that I can look at?
- What will I have at the end of the first month, in one sentence?
- Which of our systems will you connect, and which stay out?
- Where does our data go, does the vendor train on it, and how do you handle the Privacy Act 2020?
- What must a person approve before anything is sent, paid or filed?
- What is in scope, what is out, and how do you charge? Get this in writing. Compare scopes, not a day rate with no deliverable attached.
- What does my team have to do, and can we run this after you leave?
- What happens if the first build does not work?
This guide does not list prices. Ask each firm for a written scope and compare those.
What engagement types do AI consultancies offer?
Expect one of five shapes: a workshop, a done-for-you setup, a project build, an embedded engineer, or help hiring a permanent one. Buy the shape that matches the job.
- A workshop. Your people learn to do it. Fit when the bottleneck is skill, not software. Harkness AI runs this as its own practice: AI training and in-house workshops.
- A done-for-you setup. A tool connected to the systems you already run, then handed over. See Claude setup, ChatGPT setup and Grok Bot setup. Grok Bot is for teams of 1 to 5. There is no published price for it yet.
- A project build. One problem, one system: automation, a chatbot, or a custom tool. On this site a build is described as 2 to 12 weeks after the proposal. See AI automation, custom AI and Claude connected to Xero.
- An embedded engineer. Someone inside your team for part of the week, after a paid scope. Harkness AI offers this for businesses of roughly 20 to 100 staff with no internal AI function: one to three days a week, with Blake Harkness as engagement lead. Embedded AI engineer. There is no published rate.
- Help to hire. A shortlist when you want a permanent engineer. Harkness AI puts forward up to three people. The finder’s fee is flat, paid when they start, and the amount is not published. Same page, full-time hires.
A discovery call is a conversation, not an engagement. Harkness AI’s is 30 minutes and free. The sequence published on how it works is that call, an assessment (1 to 2 weeks), a proposal (1 week), the build, training, then support if you want it. Another firm will name different steps. Ask for theirs in writing. The services overview is on AI consulting for NZ businesses.
What are the red flags when hiring an AI consultant?
Walk away from a guaranteed return, a proposal written before they have seen the work, and a plan where nobody is named as the person who builds it.
- A promised return, a headcount cut, or a claim that it will run itself, with no baseline from your own numbers.
- A proposal, or a price, before they have watched the work.
- No named person for delivery. “The team” is not a name.
- Nothing already live that you can look at.
- Admin access to email, Xero or the bank before there is a written scope.
- A long programme with no deliverable in the first month.
- No answer on what the AI may send, and what a person must approve.
- Pressure to sign on the first call.
A freelancer, a boutique and a large IT firm can each show these. Judge the behaviour, not the size of the firm.
Should you hire a large IT firm, a boutique or a freelancer?
Hire a large IT firm when AI sits inside a wider technology programme. Hire a boutique specialist when you want a working system in the tools you already run. Hire a freelancer when the job is small, clear, and one person can finish it. The table is a buying guide, not a ranking.
| Large IT firm | Boutique specialist | Freelancer | |
|---|---|---|---|
| Best fit | AI is one part of a wider technology programme | A working system in the tools you already run | A small, clear job one person can finish |
| Who does the work | A delivery team, beside the firm's other IT work | A small team. Ask if the people who scope it also build it | One person |
| What you are buying | Breadth across AI and the rest of the technology estate | Depth in AI setup, training or a build | One person's time on a tight brief |
| Ask before you sign | Which person is named on your job, and what ships in the first month | Whether the specialty matches the job, and what happens if it grows | What happens when that person is away, and whether one skill set is enough |
| Handover | Who stays after the project team rolls off | Logins in your name, and a teammate who has used the system | What you can still run if that person is unavailable |
Size is not the decision. A large firm is the right call when security, identity and several systems have to move together. A freelancer is the right call for a single automation you can describe in a paragraph. A boutique sits between them: more than one person, and a focus on AI rather than the whole IT estate.
Where does Harkness AI fit?
Harkness AI is a boutique AI consultancy in Christchurch. It sets up Claude, ChatGPT and Grok Bot, trains teams, connects Claude to Xero and builds custom AI for businesses across New Zealand and Australia.
It has worked with 50+ businesses and the team has built 60+ automations. It was founded in 2025 by Blake Harkness, a Massey University mechatronics graduate. He is an AI Forum NZ AEC Working Group contributor and organiser of its webinar series, an Engineering NZ AI Advisory Group member, and the founder of Young Kiwis in AI. He was a Rising AI Star finalist at the Aotearoa AI Awards in 2025 and again in 2026. Harkness AI has been an OpenAI Select Partner since August 2026. Published work is on the case studies page.
- Done-for-you Claude, ChatGPT and Grok Bot setup, then a handover.
- Training and workshops, so a team can do the work itself.
- Automation, custom AI and Claude connected to Xero.
- An embedded AI engineer for businesses of roughly 20 to 100 staff, one to three days a week after a paid scope. No published rate.
On site for workshops and discovery in Christchurch, Auckland and Wellington, and remote for delivery elsewhere. The first step is a free 30-minute call on the contact page. There is no published day rate.
