What can AI realistically do for a small business?

Mostly the unglamorous work: drafting quotes, proposals and client emails, summarising meetings and calls, turning rough notes into process docs and FAQs, and automating routine follow-ups. AI is strongest with language and repetition, and weakest where a wrong answer really costs you — so it speeds up admin and communication rather than replacing your judgement.

What's the best AI tool for a small NZ business to start with?

For most businesses, a general assistant like ChatGPT or Claude. One subscription covers writing, summarising, drafting and analysis across the whole business, so you get value fast. Start there, get a genuine win, then add a transcription tool and a simple automation tool once that first habit has stuck.

How much should a small business spend to get started?

You can start properly for under NZ$100 a month — a general assistant is around NZ$35, transcription is free to about NZ$25, and a basic automation tool is around NZ$15. The bigger cost is the few hours of setup to build prompts and workflows around your business, and that's a one-off, not a monthly bill.

Do I need a technical person or a developer to start?

No. The highest-value starting points — writing, summarising, follow-up automation and meeting notes — don't need any code. You need someone willing to spend a few hours setting up the prompts and templates properly, not a developer.

How long before AI actually saves time?

For a writing or summarising task, the first week. The hold-up is rarely the tool — it's whether the team uses it consistently. Give each new use case about four weeks, not because setup takes that long, but because that's how long it takes to be sure the habit's stuck and the output's reliable.

Where to Actually Start With AI in Your NZ Business

The short answer: start with one high-frequency, low-stakes, text-based task — usually quote and proposal writing, client emails, or meeting notes — using a general AI assistant like ChatGPT or Claude (about NZ$35 a month). Get one real win inside a week before you add anything else. The rest of this guide is how to choose that first task and make it stick.

Nearly every owner I talk to has already decided AI matters. They've poked at ChatGPT once or twice, they've heard their competitors are "using AI", and they've got a quiet, nagging sense they should be doing more. What they haven't got is an answer to a much simpler question: start with what, exactly?

The 2026 Shaping Business Study landed in the same place. Kiwi businesses aren't short on awareness or motivation any more — they're short on the confidence and the know-how to actually put AI to work. The gap isn't belief. It's the first concrete step.

So that's all this is: the first step. No tour of forty tools, no think-piece about the future of work — just a straight answer to where to begin, and how to get something real back inside a month.

The mistake almost everyone makes first

Left to their own devices, most businesses start in one of two wrong places.

The first is the flashiest thing they can find. A customer-facing chatbot. An "AI agent" that's supposedly going to run a whole department. Whatever looked impressive in someone's LinkedIn post. These are the hardest things to get right, the most embarrassing when they break, and the slowest to pay you back. Start there and you'll almost certainly walk away deciding AI "doesn't really work for a business like ours".

The second is treating AI like a project — a big initiative with a budget, a working group and a six-month timeline. Frame it that way and it'll never ship. The adoption that actually works looks nothing like a project. It looks like one person quietly saving themselves two hours a week, then leaning over and showing the person next to them.

The right place to start is the opposite of flashy. It's the boring, repetitive, wordy work nobody enjoys and everybody does anyway.

The filter that tells you where to start

You don't need to think about AI in the abstract. You just need to find the one task in your business that ticks three boxes. Run your week through these and the starting point usually picks itself.

Condition 1

It happens often

Frequency is where the payback lives. A task you do five times a day is a far better place to start than one you do once a quarter — even when the quarterly one feels more important. Do it often and the minutes you save stack up fast, and your team gets enough goes at it to actually build the habit.

Ask yourselfWhat do my people do every single day that looks roughly the same each time?

Condition 2

It's mostly words, and low-stakes

Today's AI is at its best with language — drafting, summarising, rewriting, pulling the key points out of something messy. It's at its worst, and riskiest, anywhere a wrong answer really costs you. So your ideal first task is one where the AI produces a draft and a human still casts an eye over it before it goes anywhere. You want a head start with a person in the loop, not something you've got to trust blind.

Ask yourselfIf the output were a bit off, would someone catch it before it mattered? If yes, it's a safe place to start.

Condition 3

You already do it the same way each time

If a task has a rough shape to it — a standard quote letter, a weekly update, a particular kind of email — AI can match it. If it's pure improvisation with no pattern, it's a poor first pick. The more repeatable it already is, the faster AI makes it.

Ask yourselfCould I explain how we do this to a new hire in five minutes? If so, I can explain it to an AI too.

Whatever scores highest on all three is your starting point. For most businesses your size, that's quoting and proposals first, then client communication, then internal documentation.

Five places to start, fastest payback first

If you'd rather not run the filter yourself, here are the five spots that pay back quickest — roughly in the order I'd tackle them. Pick one. Not five.

Start 1

Writing and communication

Quote cover letters, proposals, client emails, variation notices — the same wording you've typed a hundred times. A general assistant like ChatGPT or Claude (around NZ$35 a month) turns a 25-minute job into a 4-minute one, once you've built a few prompts around your business's language and your usual job types. It's the fastest win on this list and the right first move for almost everyone.

The fixWrite five to eight reusable prompts that include your business name, your standard job types and your tone, and save them somewhere the team can actually reach. Generic prompts give you generic output — that's the difference between a tool people use and one they try twice and forget.

Start 2

Meeting and call notes

A transcription tool like Otter.ai or Fathom (free, up to about NZ$25 a month) records your site briefings, client meetings and internal calls, then writes up a summary and an action list on its own. The "write up the meeting" job just stops existing. Barely any setup, and you feel it straight away.

The fixTurn it on for one recurring meeting this week. Let the AI draft the summary and the actions — whoever runs the meeting just reviews and sends.

Start 3

Getting what's in people's heads onto paper

Most owner-led businesses run on knowledge that lives in one or two people. AI is surprisingly good at helping you get it out: hand it a messy voice note or a rough transcript of how you do something, and it'll turn it into a clean process doc, a checklist or an FAQ answer. This is the quiet one that makes the business a little less dependent on you — which, let's be honest, is the whole game.

The fixRecord yourself explaining one process out loud for three minutes. Ask the AI to turn it into a step-by-step procedure. Edit, save, done — you've just written up something you've meant to get to for a year.

Start 4

Joining up the tools you already use

An automation tool like Make.com or Zapier (from around NZ$15 a month) connects your existing systems so the routine sequences run themselves. The big one for most businesses is follow-up: a quote goes out, and if there's no reply in three days a reminder sends itself; if there's still nothing five days later, the estimator gets a nudge to pick up the phone. A bit more technical than the first three, but the payback's high.

The fixAutomate exactly one sequence first — quote follow-up is usually the winner. Get it running reliably before you build a second.

Start 5

Marketing and content

Drafting social posts, newsletters, website copy, first drafts of articles. Genuinely useful and low-risk — but it's last on purpose. It rarely touches your biggest time drain, and it's the easiest one to disappear into, fiddling instead of shipping. Get an operational win under your belt first, then come back here.

The fixUse it to draft, never to publish as-is. Your voice and your judgement are the value; the AI's just there to kill the blank page.

The businesses that get real value from AI don't start with the most impressive use case. They start with the most repetitive one — and let the wins build from there.

The bit nobody mentions

Sort the process first

If a task is undocumented and inconsistent — three people do it three different ways and nobody's agreed on a standard — AI won't fix that. It'll just produce inconsistent output faster. A prompt only works when it's built around a process that actually exists. So sort the process, then automate it. That's why workflow documentation comes before AI in every job we run, and it's the single biggest reason DIY efforts stall.

What not to start with

A quick list of things that look like good starting points and aren't — at least not yet.

Autonomous, customer-facing AI. A bot answering customers with no human in the loop is the highest-stakes, hardest-to-control thing you can build. It comes much later, if at all.

Anything high-consequence. Legal, tax, financial, health and safety, hiring and firing. AI can help you draft questions or get your thinking in order, but the judgement — and the liability — stay with a human, and a qualified one where it counts.

Anything that replaces judgement instead of admin. The point at the start is to take the typing, formatting and chasing off your team's plate, not to hand over the thinking that makes you good at what you do.

One rule before anyone starts: your data

Treat consumer AI tools like a public-facing draft, not a vault. Don't paste client personal details, financials or anything covered by the Privacy Act into a consumer plan without checking the data settings first. Most business and team plans let you switch off training on your data — use those, and write one plain line for the team about what can and can't go into a tool. Five minutes now saves you a real headache later.

A 30-day plan to actually start

You don't need a strategy. You need a month and one use case.

Week 1 — Pick one. Run the three-box filter, or just take Start 1. Choose the task with the most frequency and the least risk, and resist the urge to grab three.

Week 2 — Build it properly. Spend two or three hours writing the prompts or setting up the workflow around your business. This is where the value's made or lost, so don't rush it. Save everything somewhere the team can find it.

Week 3 — Make it the way it's done. Run a 30-minute session with the two or three people who'll use it most. It's not an optional extra they reach for when they remember — from now on, it's just how that task gets done.

Week 4 — Check it stuck. Is the team actually using it? Is the output reliable? If yes, then — and only then — pick the second use case. If not, fix the friction before you add anything.

The honest economics: a starter kit of a general assistant, a transcription tool and one automation runs comfortably under NZ$100 a month. The real cost is six to eight hours of setup, paid once. Against that, getting back even five hours a week — at a conservative $45 an hour loaded — is around $1,000 a month of capacity you didn't have before. That's the maths of AI done properly, and it's why starting small beats starting impressive.

Want the version built for a specific trade? Our breakdown of three AI tools that save a NZ construction business 12 hours a week walks through the exact tools and workflows. And if part of what's pushing you toward AI is a feeling that the business has quietly stopped scaling, the signs your business has hit a ceiling covers the ground around it.