How to choose an AI consultant in New Zealand

An AI consultant worth hiring in New Zealand will start from one measurable business problem, show you working systems the firm has built, explain in writing how your data is handled under the Privacy Act 2020, and propose a small paid first stage with a defined deliverable. Before you sign anything, ask a Regional Business Partner growth advisor whether the work qualifies for government co-funding, because eligible small businesses can claim up to half of eligible costs for an AI roadmap and implementation support.

What an AI consultant actually does

"AI consultant" covers several kinds of firm. Some advisors assess your business, write a roadmap and hand the build to someone else. Implementation firms write code, connect systems and run the result in production. Software resellers and platform partners increasingly offer AI advice as well, usually centred on the product they sell, and plenty of firms mix these roles.

The mix matters because a firm's advice tends to follow its revenue. A platform partner earns more when you adopt the platform, while a strategy-only advisor carries no risk if the roadmap proves too expensive to build. A firm that builds software has its own reason to recommend a build. Any of these models can serve you well once you know which one you're dealing with, so ask directly how the firm makes its money on a typical project.

For most small and mid-sized New Zealand organisations, useful AI work clusters around a handful of areas: extracting data from documents such as invoices, orders and claims; triaging customer emails and enquiries; searching internal policies and records in plain language; drafting routine correspondence and reports; and forecasting demand or cash flow from existing records. A competent consultant spends the early meetings on your processes, systems and data quality before discussing models or tools.

Do this before you contact anyone

Pick one process that costs you real time or money and measure how the process runs today, whether in hours per week, error rate, turnaround time or cost per transaction. Without a baseline, nobody can show you whether a project worked. List the systems involved (accounting, CRM, ERP, shared drives), note where the data lives and who controls access, and name one person in your organisation who'll own the project and can make decisions about it.

The Government's Responsible AI Guidance for Businesses, released in July 2025 alongside New Zealand's first AI Strategy, recommends starting with a manageable project tied to a real business problem and choosing providers that meet New Zealand's legal and ethical standards. The guidance is voluntary, though following its advice on starting small will make every consultant conversation more productive.

Questions to ask an AI consultant

About the people doing the work

Ask who will do the work, by name, and whether the people in the sales meeting will stay on the project. Ask to see a system the consultant built that's running in production, shown live with the client's permission, or a working demo on data similar to yours. Slide decks and generic chatbot demos tell you very little about delivery. Ask what went wrong on a recent project and what the firm changed afterwards, since a consultant with real delivery experience will give a specific answer.

About your data and systems

Ask which AI models and providers the consultant proposes, where those services process and store data, and whether your prompts and documents could be retained or used to train someone else's model. Find out how the solution connects to your existing systems and who maintains those connections when a vendor changes its software. Ask what accuracy the consultant expects, how errors will be caught and which decisions will stay with a person.

About cost and ownership

Ask for a fixed price on the first stage and a written estimate of running costs afterwards, including model usage fees, hosting, licences and support. Agree who owns the code, prompts, configuration and any trained models when the engagement ends, and how you'd move the work to another provider. A consultant who can't describe an exit path is asking you to accept lock-in.

Warning signs

Be wary of a consultant who promises percentage savings before seeing your processes or data, or who names a preferred platform in the first meeting before understanding the problem. Requests for a long retainer up front, vague answers about where data goes and silence on staff training or change management all deserve a hard follow-up question.

Treat claims of official approval with care. The Office of the Privacy Commissioner states that it doesn't approve or endorse any AI technology or AI technology providers, so a consultant claiming the Commissioner's backing is misrepresenting the Office's position. A directory listing isn't a quality mark either. The AI Forum New Zealand's AI Capability directory is designed to help organisations find AI-related support and services, and the AI Forum says inclusion in the map doesn't imply an endorsement.

What a good first engagement looks like

A sound first engagement is small, paid and fixed in price, and it ends with something you can use whether or not you hire the same firm again. In most cases the engagement opens with a discovery or roadmap stage lasting a few weeks. The output should include a prioritised list of opportunities with estimated effort, cost and benefit for each, an honest view of whether your data is ready, the privacy risks involved and a recommended pilot with success measures tied to the baseline you recorded.

The pilot comes next, covering one process with real data, a limited group of users and human review of every output. At the end of the pilot you compare results with the baseline and decide whether to expand the work or stop. A good consultant treats stopping as a legitimate outcome and designs the pilot so the numbers make the decision obvious. If a proposal jumps straight to a large implementation, or runs an open-ended discovery phase with no defined output, ask for the work to be split into stages with a decision point between each.

How to check a consultant's claims

Ask for two references from clients of similar size or with similar systems, and call the references yourself. Useful questions include what the consultant proposed compared with what the firm delivered, whether the system is still in use six months later and how the firm handled problems. When a consultant quotes a result such as hours saved, ask how the figure was measured and over what period.

Funded work gives you an extra check. Training paid for through the Management Capability Development Fund must be approved management training from registered service providers, and you work with a Regional Business Partner advisor to identify training needs and connect with approved providers. Talking a consultant's proposal through with a growth advisor is worthwhile even when funding isn't your main goal.

Data and privacy questions to settle

The Privacy Act 2020's information privacy principles apply whenever an AI tool handles personal information. The Privacy Commissioner's AI guidance, created in September 2023, recommends doing a Privacy Impact Assessment before you start and updating the assessment regularly. The same guidance says you must keep personal information secure, including prompts and training data, and recommends that if you're in doubt, you don't use AI tools to handle personal information.

Offshore processing needs particular attention, because many AI services run on overseas infrastructure. Information Privacy Principle 12 allows personal information to be disclosed to an overseas recipient only in set circumstances, such as when the recipient is subject to the Privacy Act because it does business in New Zealand, or is bound by privacy laws or contract terms that provide comparable safeguards. Where none of those circumstances apply, disclosure needs the individual's permission after the person has been expressly told the information may not receive the same protection. A newer rule affects AI tools that gather information about people from third-party sources: IPP3A came into force on 1 May 2026 and requires an agency collecting personal information indirectly to take reasonable steps to make the individual aware, unless an exception applies.

Get written answers on where data is processed and stored, how long the provider retains data, whether data can be used for model training, who at the consultancy can access your data, how breaches will be reported to you and how data is returned or deleted when the contract ends. If those answers leave you uneasy, price a private deployment that keeps data on your own network.

Funding help in New Zealand

Both programmes run through the Regional Business Partner Network, so you register your business with the network and speak with a growth advisor. The AI Advisory Pilot is administered by MBIE, and because places are competitive, eligibility alone doesn't guarantee funding. The pilot excludes several kinds of cost, including developing AI products or services to sell, general software or product development, costs outside the agreed scope and work done before approval. The exclusion for work done before approval has a practical consequence: don't let a consultant start until the funding decision is confirmed, or you may pay the full cost of that work yourself.

Where DataForge fits

DataForge is an Auckland consultancy based in Hobsonville, working on AI consulting, automation, custom software, ERP and IT service management for New Zealand and Australian organisations. We build what we recommend, so the team that scopes a project also delivers the project. For clients who'd rather sensitive data never left their premises, we can set up a private AI server so data stays on the client's network. We also make our own products, POS Forge and ForgeBox, and you should expect us (as you'd expect any consultant) to tell you plainly when a recommendation involves our own software. We quote prices after a free assessment, and our AI consultancy page explains how we work.

Checklist for choosing an AI consultant

• One process chosen, with a measured baseline
• Systems, data locations and data owners listed
• Internal project owner named
• Consultant has named the people who'll do the work
• Production system or realistic demo seen live
• Two references called
• AI models, providers and data locations confirmed in writing
• Prompt retention and training use answered in writing
• Any offshore disclosure checked against IPP12
• Privacy Impact Assessment planned before the pilot
• Fixed price agreed for the first stage
• Ongoing running costs estimated in writing
• Ownership of code, prompts and models agreed
• Exit, data return and deletion process agreed
• Growth advisor contacted and any funding confirmed before work starts

Your next step

If you'd like a starting point for your own shortlist, two free options from us cost nothing but a few minutes, and the funding step applies whichever consultant you choose.

• Take the free AI Opportunity Score, a 2 minute quiz.
• Book a free 30 minute walkthrough with the DataForge team to talk through your process and data.
• Register with the Regional Business Partner Network and speak with a growth advisor before any consultant starts work.

More guides

• AI consulting in NZ: what it costs in 2026
• AI consultants in New Zealand: a 2026 guide
• AI consultants for New Zealand businesses: how DataForge works