You are reading this on the blog of an AI implementation firm, and we have put ourselves at the top of the list. So this is not an independent review, and you should not read it as one.
What I can offer instead is specificity. We implement Claude for regulated financial firms for a living, two of my cofounders and I were part of Trineo, one of New Zealand's most successful cloud-platform implementation partners through the SaaS wave, and between those two things I have watched a lot of technology projects land badly in this market. So I know what these engagements actually look like from the inside. Every entry below says what the firm is good at and the kind of buyer it suits, drawn from what each one publishes about its own work. Only one entry has a section on where the firm falls short, and that is ours, because it is the only one on this list I know from the inside. I am not going to pass judgement on firms I haven't worked alongside.
The scope is narrow on purpose: firms doing AI implementation work with New Zealand financial services. Insurers, brokers, lenders, wealth and advice businesses, and the accounting and legal firms that sit alongside them. That excludes a number of capable general AI shops, and it includes the Big Four, whose advisory practices are usually already in the room.
What this list is ranking
Almost every "best AI companies in New Zealand" page you will find is a directory scrape with the same eleven logos in a different order. Ranking by headcount or by review count tells you nothing about whether a firm can help a FAP-licensed broker put a model near a client file.
So I ranked on two things. First, how much of the firm's delivery work sits in regulated financial services. Second, whether what they leave behind can be shown to somebody who asks a hard question about it later, whether that is your own board or the FMA. A working automation that nobody can explain six months on is a liability wearing the costume of a win.
1. Airclerk
We are a small firm doing one thing: getting Claude into regulated financial and professional services in a way that survives contact with a regulator. That means scoping the work properly to decide which processes should go anywhere near a model, workflow implementation for the ones that should, MCPs and connectors so the assistant can see your actual systems rather than a copy-paste of them, and AI governance that a licensed person can sign.
The part we care most about is AI Process Assurance: a business-readable record of the AI-assisted work you choose to instrument, so what the model did, how it was checked and who signed off is retained rather than reconstructed. Most AI work in this market produces output. Very little of it produces evidence.
We also build the vertical depth ourselves rather than buying it in. There are Claude skill packs for New Zealand accounting and New Zealand law, written against local rules, and a Xero command line tool and MCP built in-house because the general-purpose integrations weren't good enough for year-end work. Our view of why the industry knowledge matters more than the tooling is here.
The honest bit. Airclerk is young. The Trineo history is real and the three of us have decades between us, but the company itself does not have a fifteen-year client list, and if your procurement process scores on vendor longevity we will lose that box. We are also opinionated to a fault: we implement Claude, so if you have already standardised on a different model, we are not the neutral advisor you want. We are the wrong firm for marketing automation, or anything where the goal is volume rather than defensibility. And the assurance work only pays for itself where the process genuinely matters. If you want to see whether AI can tidy up your internal meeting notes, hire somebody cheaper.
2. The Big Four: PwC, Deloitte and EY New Zealand
The large advisory firms all run AI practices in New Zealand now. PwC New Zealand publishes an AI services line covering implementation and oversight, including automating high-volume work like claims processing. Deloitte New Zealand has been writing about AI in banking and insurance for a while and has the sector practices to match.
Their genuine advantage is not the technology. It is that they already know your business, they can put twenty people on something next month, and a partner's signature carries weight in a board pack in a way a five-person firm's does not. For a programme spanning eight business units and a core system replacement, that coordinating capacity is worth what it costs.
3. Arkham Solutions
Arkham is an AI automation agency serving New Zealand and Australian businesses, and they are refreshingly unambiguous about how they operate. Every engagement runs audit, then build, then deploy and improve, over 30 to 60 days. The first audit is a free scoping call where they map manual workflows and score them by return. The build phase assigns a senior architect by name, with prototypes inside a fortnight. Then they monitor for 30 days and you can cancel any time on 30 days' notice.
That shape is well suited to a broking or lending operation drowning in reconciliation, reporting, compliance returns and client correspondence. They name finance among the sectors they work in, alongside insurance, professional services and healthcare. They describe their client base as running from Auckland trades to Australian enterprise, which tells you the model travels.
The free first audit is worth calling out on its own. Plenty of firms in this market will not tell you where the opportunities are until you have signed something, and a scoping call that hands you a written plan is a genuinely low-risk way to find out what your own manual work is costing you.
4. AI Surge
AI Surge describe themselves as Australia and New Zealand's full-cycle Claude AI partner, and they are the closest thing to a direct competitor on this list. Their material covers the whole Claude surface properly: Claude Chat, Claude Cowork for integrated workflows, and Claude Code and agents for autonomous work. They sell strategy, two to four week implementations, team training, workflow automation and MCP connector development, and they say governance and compliance are built in from day one.
They also list wealth and financial advisory explicitly, and say they have deployed Claude for legal, accounting and consulting firms. If you want Claude rolled out across finance, HR, legal and operations at pace, and you would rather one firm handled the lot, they are a real option and they are clearly doing volume.
Their range is wide: manufacturing, mining, construction, agriculture, healthcare, property, shipping, retail and education sit alongside financial advisory. Having deployed the same tool into that many operating models teaches you things a specialist never sees, and if your firm does not think of itself as a typical financial services business, that breadth may fit you better than a narrow practice would.
5. Binary Refinery
Binary Refinery is an independent AI strategy and governance consultancy, and independence is the whole point of them. They work on AI resilience and disruption risk, capability and governance, technology transformation, digital due diligence for M&A, and workforce readiness. Half-day workshops are published at $4,500 + GST, which is more pricing transparency than most of this market manages.
They pay real attention to regulated professional work. Their writing covers what Claude's expansion into legal actually changes for New Zealand law firms, which is a more useful question than most commentary on the subject bothers with. For a board that wants an honest assessment before committing budget, an advisor with nothing to sell you afterwards is worth a lot. They are also down the road from us, and they are good.
Their engagement ends with a decision rather than a build, and that is the design, not a gap in it. An advisor who will not be quoting on the implementation can tell you to do less, or to wait, which is advice the rest of us are structurally worse placed to give.
6. Advancer
Advancer is a technology company that bridges strategic consulting with product development, and they say they started as practitioners frustrated by the gap between AI hype and AI reality. Their work covers AI strategy, board and executive-level technology leadership, and building production AI systems, and they run their own platforms including ThinkSpace, SupportSpace and Aiqbee. "Substance over hype" and only recommending what they know works is the stated position, and their material reads like engineers wrote it.
They are the pick when the problem is genuinely technical. A retrieval pipeline over twenty years of policy documents that has to return the right clause, a legacy integration everyone else has bounced off, or a team that needs senior technical leadership for six months without hiring a permanent CTO.
They suit a firm with its own risk and compliance function already in place, where what you are missing is people who can build the thing properly. That division of labour works well, and it is how a lot of the better in-house AI systems in this market have actually been delivered.
The option nobody quotes you for
Doing it yourself deserves to be on this list, because for a decent number of firms it is the right answer and no consultancy will volunteer it.
If you already have a platform licence, one person internally who is genuinely good at this, and an appetite to learn slowly, start there. Pick a process that matters but won't hurt anyone if it goes sideways, and have your own people build it. You will learn more in six weeks of that than in any strategy engagement, and the knowledge stays in the building. We wrote about why owning the loop beats renting it.
The reason to bring somebody in is not that AI is hard to switch on. It is easy to switch on, which is exactly the problem. You bring someone in when the work touches a licensed obligation or a number that ends up in a return, and being wrong is expensive in a way that is not recoverable by trying again. For everything below that line, your own team and a decent licence will do.
If the useful thing right now is a conversation at that level rather than a project, that is what AI Platform Advisory is for.
How to choose, in five questions
Whoever you shortlist, these are the questions I would put to them. They are the ones we get asked by the sharpest buyers, and the ones I would want asked of us.
- Who signs off on what the model produces, and can that person explain it? If the answer involves nobody in particular, you have bought a risk rather than a capability. Regulated work needs a named human at the end of it.
- Am I buying a plan or a working system? Get the deliverable named in writing, with the date it goes live. "Roadmap" and "in production" are very different purchases at surprisingly similar prices.
- What happens when it gets something wrong? Not whether it will. Ask how the failure gets noticed, who it escalates to, and what the record of it looks like afterwards.
- Do they know which regulator applies to me without checking? A firm that works in your sector knows whether CoFI, CPS 230 or the Contracts of Insurance Act is the thing shaping your next eighteen months. One that doesn't will be learning your compliance obligations on your budget.
- Who does the work, and are they named in the proposal? Ask for the actual people and how much of their week you get. This single question separates most good engagements from most disappointing ones.
The uncomfortable part of this market is that the technology is now the easy bit. Any of the six firms above can get a model running against your data inside a month. The question worth putting to all of us is what gets left behind: whether the thing still works when the person who championed it moves on, and whether you can show your board how it reached the answer it reached. That has always been the harder half, and it was never really a technology problem.
If you think we're the right fit, get in touch. If one of the other five suits what you're doing better, they are all linked above, and I would genuinely rather you talked to them than hired us for work we're not the right shape for.