AI in New Zealand: what actually works

There are two versions of AI going around. The one on stage: agents running companies, jobs gone, a wave you’ve already missed. And the one in the back office, where someone is quietly trying to work out if any of it is worth the money. The second one is the real conversation. Here’s where it’s actually at.

What works

The stuff that works is boring. That’s the point. High volume, low stakes, mostly words.

• Drafting. Emails, quotes, proposals, job ads. Not the final version. A first pass someone then fixes.
• Summarising and searching. Making a pile of documents or a full inbox usable again.
• First-line support. The same twenty questions your team answers every week, handled, and passed to a human when it gets out of its depth.
• Coding. If you have developers, this one mostly lives up to the hype.

Same thread through all of them. It speeds up work you already do, a person stays in the loop, and a wrong answer costs nothing. That’s a good first project. Small, useful, hard to get badly wrong.

What doesn’t

The bigger the claim, the worse the odds. “Replaces your team.” “Runs itself.” “Build your own model or fall behind.” Good headlines. Bad plans.

The wins are real. They’re just smaller and quieter than the pitch. Most of what gets promised never ships.

The tech is fine. The problem is the distance between a slick demo and something your team can rely on every day. That distance is where the money and the disappointment go.

The famous tool probably isn’t yours

The best known tool is usually just the best marketed one. What fits you depends on your data, your people, your appetite for risk and your budget. None of that matches the company in the case study. “Everyone’s using it” is a reason to look. It’s not a reason to buy.

The New Zealand bit

A few things are different here. Small market, so a lot of tools are built for someone else’s rules and someone else’s scale, and local support is thin. Margins are tight, so a six figure “transformation” is rarely the first move that makes sense. And plenty of businesses here care where their data ends up, which is fair enough.

Being small has an upside. You can go slow on purpose. Pick one annoying job, try something cheap, see if it holds up before you spend real money.

Where to start

• Pick a boring problem that already costs you time. Not “use AI.” More like “stop rewriting the same five emails.”
• Do the maths. Cost, saving, time to get there, what breaks if it goes wrong. If it doesn’t add up, that’s your answer.
• Keep a person in the loop, at least early on.
• Check back in a month. Are people still using it, or did it quietly die.

The point

AI isn’t magic and it isn’t a scam. For most businesses here it’s a handful of genuinely useful tools that, picked well, take real work off your plate. Picked because it was in the news, it’s an expensive lesson.

Working out which is which is the whole job.