If you run a business in Aotearoa New Zealand, you’ve probably had at least one person on your team bring up AI in the last six months. Maybe more than one. And the conversation usually lands in the same place, which is a quiet mix of excitement and worry. Excitement about the time your team could get back. Worry about whether the tools are safe to use, whether anyone is going to get left behind, and whether you should be doing more.

This article is for the owner who is ready to take that conversation seriously without falling for the noise. I’ll walk through what AI upskilling actually looks like for a Kiwi business in 2026, what it costs in plain NZD, where the legal landmines sit, and how to put together a plan your team will actually follow through on.

The State of AI in NZ Workplaces Right Now

We work with NZ and Australian businesses every week on this, and the picture in 2026 is uneven. The big corporates and the well-funded tech firms have been training staff on AI for over a year now. A lot of smaller Kiwi businesses, the ones with five to fifty people, are still figuring out where to start. Some have staff quietly using ChatGPT on their personal logins. Some have banned it outright. Most are somewhere in between.

Industry estimates suggest around a third of small NZ businesses have any formal AI policy at all. That number is rising, but slowly. What we typically see in our network is that the businesses making real progress are the ones treating AI upskilling as a normal workforce capability, like cybersecurity training or health and safety induction, rather than a special project.

The other thing worth saying upfront is that this is not a “wait and see” moment anymore. The tools are already in your workflow. Xero has AI features rolled into its plans. MYOB has built automation into the bank reconciliation and coding side. Trade Me, Seek, and REA Group all use machine learning to rank listings and applications. If your team isn’t using these features deliberately, they’re using them accidentally, which is usually worse.

What “Upskilling” Actually Means in 2026

A common mistake we see is treating AI training like a one-off course. Someone sends the team through a two-hour webinar, everyone gets a certificate, and nothing changes. That approach has been failing for decades on every other piece of software and it fails here too.

Upskilling in 2026 looks more like three layers stacked on top of each other. The first layer is literacy. Every person in your business should be able to explain in plain English what an AI model is, what a large language model does, and what it is bad at. Not the technical detail, just the working knowledge. The second layer is tool-specific skill. How do you actually use Copilot in Word, or the AI features in Xero, or the prompts that work in ChatGPT for your specific job. The third layer is judgement. When should you trust the output, when should you double-check it, and when should you not use it at all.

Most businesses skip straight to layer two and never come back for layers one and three. That’s why the same mistakes keep happening. Someone uses AI to write a customer email and ships it without reading it. Someone pastes client data into a public chatbot. Someone trusts a model summary of a contract clause that turns out to be completely wrong.

The Skills Your Team Genuinely Needs

The job-specific skills will vary. An accountant needs different AI skills than a builder or a marketing coordinator. But there is a shared core that everyone benefits from, and it’s worth training to that level first.

Prompting is the obvious one, and most people are worse at it than they think. The difference between a useless prompt and a useful one is usually about specificity, context, and the format you ask for. A good training session will cover the patterns, not just the theory. We typically see staff improve their output quality dramatically within a week once they get the basic patterns down.

Data hygiene is the underrated one. AI tools are only as good as the data and context you feed them. If your team doesn’t know what good inputs look like, the output will be unreliable. This is especially true for the businesses we work with who are feeding AI their own internal documents. If those documents are a mess, the AI output is a mess.

Critical evaluation is the third core skill, and it’s the one most often missing. Your team needs to be able to read an AI output and ask themselves whether it actually answers the question, whether the numbers add up, and whether anything important is missing. This is where most of the risk lives.

The fourth skill, which matters more in 2026 than it did a year ago, is workflow design. Knowing how to slot AI into a process so it actually saves time, rather than adding another step. One Auckland accountant in our network rebuilt their monthly close process around AI assistance and cut the team’s after-hours work by roughly a third. The change wasn’t the tool, it was the redesigned process.

Privacy, Compliance, and the Stuff That Can Bite You

This is where NZ businesses need to slow down. The NZ Privacy Act 2020 is the framework that governs how you handle personal information, and its 13 Privacy Principles apply as much to AI tools as they do to your filing cabinet. Principle 12 in particular covers offshore disclosure of personal information, and a lot of the AI tools your team wants to use are hosted offshore. That is not automatically a problem, but it is something you need to think about.

The short version is this. If your team is pasting client names, email addresses, IRD numbers, or any other personal information into a public AI tool, you may already be in breach of PP12, depending on what that tool does with the data and what its terms say. Some tools offer enterprise tiers with data residency and contractual protections. Most free or cheap consumer tools do not.

A few practical rules we recommend to the businesses we work with. Don’t put identifiable client or customer data into consumer AI tools. Use the enterprise or business tier of any AI tool that will handle personal information. Read the data handling terms, or have your lawyer or IT advisor read them. Turn off training on your data where the setting is available. And keep a simple internal register of which tools are approved for which use cases.

If you operate in Australia as well, the equivalent frame is the Privacy Act 1988 and the Australian Privacy Principles. ASIC’s Regulatory Guide 265 on electronic trading and the use of AI in financial services is also worth a look if you’re in that sector. APRA’s CPS 234 applies if you’re an APRA-regulated entity and covers information security, which increasingly means AI security too. For healthcare, the AHPRA codes and guidelines set expectations on the use of AI in clinical and administrative work. None of these are reasons to avoid AI. They’re reasons to do it properly. Verify the specifics with your lawyer or compliance advisor, because the detail changes and your situation has its own quirks.

What a Realistic Training Plan Looks Like

A training plan that actually works in a small or mid-sized NZ business tends to follow a similar shape. It starts with leadership, not the team. If the owner or the leadership group doesn’t understand the basics, the rollout will be incoherent. We’ve seen this play out badly, with managers telling staff to “use AI” without being able to model what good use looks like.

The second step is a policy. It doesn’t need to be 40 pages. One or two pages that say which tools are approved, what data can and can’t go into them, who to ask if you’re unsure, and what to do if something goes wrong. The third step is a structured training rollout, usually over four to six weeks, with a mix of short self-paced modules and one or two live sessions. The fourth step is measurement and feedback. Without that, you’ll never know if it’s working.

For a business of 20 staff, a sensible budget in 2026 looks something like this. LinkedIn Learning or Coursera Plus subscriptions for the team will run roughly 50 to 100 NZD per person per month, depending on the tier. A dedicated live workshop with an AI training provider will typically cost between 1,500 and 5,000 NZD per session for a group of that size. An internal policy and tool register, if you get help building it, is a few thousand more. None of those numbers are precise, and pricing varies, so verify with the provider before you commit.

If you’re a smaller business with five to ten staff, the math is different. You’re probably looking at self-paced learning, one or two focused workshops, and a lot of internal coaching from whoever on your team picks it up fastest. The good news is that the free content available in 2026, including Microsoft Learn, Google Skillshop, and the public training from the major AI vendors, is genuinely good for foundational skills.

Where the Real Time Savings Come From

The businesses getting the best return on AI upskilling in our network tend to be focused on a small number of high-volume tasks. For an accounting practice, that might be drafting client emails, summarising meeting notes, and building first drafts of advisory reports. For a trade business, it might be writing job descriptions, drafting quotes, and producing maintenance instructions for clients. For a professional services firm, it might be research summaries, document review, and proposal drafting.

The pattern is the same. Pick the three to five tasks your team does most often, where the output is reasonably structured and the cost of a small error is low. Train the team specifically on those tasks. Measure the time before and after. Expand from there.

One Sydney law firm I spoke with recently has been doing exactly this, focusing on AI use for first-draft contract review and disclosure schedules. The partners made it clear that the AI is doing the boring first pass and the lawyers are still doing the substantive review. That framing matters, because it sets the right expectation across the team and keeps the humans accountable for the output.

Common Pitfalls Worth Naming

A few things to watch for. The first is letting one person become the “AI person” and everyone else outsources their thinking to them. That concentrates risk and slows down adoption. The second is buying tools before you’ve trained the team, which is one of the most expensive ways to fail. The third is measuring success by number of prompts run or licences issued, which tells you almost nothing about whether the business is actually benefiting. Measure time saved, output quality, and revenue or margin impact, in that order.

The fourth pitfall is ignoring the cultural side. Some staff will be enthusiastic. Some will be anxious. Some will be openly skeptical, and often for good reason. A training plan that doesn’t make space for those conversations usually produces surface-level compliance, not genuine capability.

What to Do This Quarter

If you’re reading this and you don’t have a plan yet, the simplest starting point is this. Pick one person on your team who is curious and reasonably competent, give them time to spend a week learning the foundations across one or two tools that are relevant to your business, and ask them to come back with a short written recommendation. Use that as the seed for your policy and your wider rollout. Total cost before you spend a cent on licences is roughly the cost of that person’s time for a week, which is usually the cheapest piece of advice you’ll ever get on AI.

From there, build the policy, set up the approved tools with the right data settings, run a short training sprint with the whole team, and pick the three to five workflows where you’ll measure the impact. Revisit in 90 days. Adjust. Keep going.

The businesses that will look back on 2026 as the year AI genuinely changed how they operate are the ones who treated it as a workforce capability, not a software purchase. That is the through-line. Train the people, set the rules, measure the result, and iterate.

A Note on Pricing and What to Budget

The figures in this article are approximate, based on the typical pricing we see in the NZ and Australian market in 2026. USD prices have been converted at roughly 1.65 NZD per USD and 1.55 AUD per USD, but provider pricing changes often, so confirm before you commit. For a business of 10 to 30 staff, a realistic first-year budget for AI upskilling, including tools, training, and external support, is somewhere between 15,000 and 60,000 NZD. That range is wide because the right number depends on how much you do in-house, which tools you choose, and how aggressive the rollout is. The lower end assumes mostly self-paced learning and a careful tool selection. The upper end assumes dedicated workshops, premium tool tiers, and external advisory support.

For smaller businesses under ten staff, the same scope can often be done for between 5,000 and 20,000 NZD in the first year, with most of the spend going to tool subscriptions and a couple of focused training sessions.

Getting the Right Help

AI upskilling is one of those areas where the generic advice is easy to find and the specific, contextual advice is what actually moves the needle. The difference between a useful rollout and an expensive one usually comes down to a few early decisions about tools, policy, and which workflows to focus on first.

Enterprise DNA works with NZ and AU businesses on this challenge. Book a 60-min Omni Audit — https://calendly.com/sam-mckay/discovery-call?utm_source=edna-landing&utm_medium=blog&utm_campaign=nzau