Almost nobody publishes real numbers for AI work. Sales pages talk about transformation and efficiency, quotes arrive with a single total on them, and business owners are left trying to budget for something they can't get a reference price on.
So here's the honest version, in New Zealand dollars, based on what I actually charge and what clients actually end up paying to run these things. Prices exclude GST. Your situation will vary, but not by as much as the vagueness implies.
The three cost layers people conflate
Almost all confusion about AI pricing comes from mixing up three separate things.
Subscriptions are the tools you rent. Predictable, small, monthly.
Build cost is the one-off price of making something specific to your business.
Running cost is what it costs to keep that thing working once it exists — the usage charges, the hosting, and the maintenance. This is the one people forget, and it's the one that turns a good project sour if nobody mentioned it up front.
Layer one: subscriptions
A business licence for one of the main assistants runs roughly $30 to $60 per person per month. For a team of five, call it $150 to $300 a month, or somewhere around $2,000 to $3,600 a year.
Worth knowing: the free consumer tiers may use what you type to improve the product. The business tiers contractually don't. If staff are pasting customer details into a free account, that's a decision someone should make deliberately rather than by accident.
My honest view on subscriptions is that they're cheap enough to be worth trying and weak enough that they rarely change a business on their own. They make individuals a bit faster at writing and thinking. They don't fix a process. If you buy nothing else, buy a couple of these and see who actually uses them after a month — that tells you something useful for very little money.
Layer two: build cost
This is where the range gets wide, so it helps to think in tiers.
A single contained job — roughly $1,000 to $3,000. One specific thing, done properly. Photographed job sheets or supplier invoices becoming structured records. An assistant that answers from your own documents. Enquiries being sorted and routed automatically. These are small, sharp projects with an obvious before and after, and they're where I send most people first.
Something a team uses daily — roughly $4,000 to $12,000. Now it's real software: multiple users, permissions, an interface people rely on, edge cases that matter because someone's job depends on it working on a wet Tuesday. A field app for staff on the road with AI handling the admin behind it sits here.
A system the business runs on — $12,000 and up. A custom CMS and CRM built around how a specific business works, with AI woven through the operational parts. Fewer businesses need this than are sold it, but when it's right it replaces several subscriptions and a lot of manual work.
The honest thing to say about that spread is that the difference between a $1,500 project and a $10,000 one is almost never the AI part. It's how many people use it, how much it has to integrate with, and what happens when it gets something wrong.
Layer three: running costs
The part that gets left out of quotes, so ask about it explicitly.
Usage charges. Most AI features bill per use. For a small business tool this is usually genuinely minor — often $10 to $80 a month, sometimes less. It scales with how much you use it, which is fine, but you want to know the shape of that curve before you sign, not after a busy month.
Hosting. Typically $0 to $50 a month for the kind of thing I build, because modern hosting is cheap and scales down as well as up.
Maintenance. The one worth thinking hardest about. Software that touches AI providers needs occasional attention, because the providers change things. Budget something for this. A care plan around $89 a month covers it for most small builds; a bigger system needs more.
So a typical small custom tool might be $2,000 to build and $60 a month to run. That second number is the one that should appear in your quote.
What makes a project expensive
In rough order of impact: the number of people who use it daily; whether it has to talk to systems you already have; whether it needs to work offline or on a phone in the field; how bad it is when it gets something wrong; and how much your existing information needs cleaning up before anything can use it.
That last one surprises people. If your records live in three spreadsheets with inconsistent names, the first part of the project is fixing that, and it isn't AI work at all.
How to know if it's worth it
Do the sum before you commit. If a task takes someone five hours a week at $35 an hour, that's about $9,100 a year. A $2,000 build that removes most of it pays for itself inside three months and then keeps paying. If a task takes twenty minutes a fortnight, no build is worth it — and I'd tell you so.
That calculation is nearly always available before you spend anything, and anyone unwilling to do it with you is telling you something.
Getting a real number for your situation
If you want the specific figure for your business rather than a range, that's what an AI Clarity Session is for: 90 minutes on your actual operation and a written one-pager covering what's worth doing, what isn't, and roughly what each would cost. It's $450, and it comes off the price of the build if you go ahead. If the honest answer is that nothing here is worth your money yet, that's what the one-pager will say.
