How to choose an AI automation agency you won't regret.
A buyer's guide for owners in New Zealand and Australia. What an AI automation agency actually sells, what the market charges, the red flags that show up before the contract does, and the twelve questions worth asking on the first call.
Most owners pick an AI automation agency the way they'd pick a printer.
You search “AI automation agency”, open five websites, book two calls and choose the one with the best demo and the friendliest quote. Three months later there is a chatbot on the website nobody uses, a monthly invoice, and a workflow that broke the week the person who built it moved on. Plenty of owners we speak to arrive brassed off with the whole category, and fair enough.
The problem is not that AI doesn't work in a services business. It does, and the good builds pay for themselves quickly. The problem is that the buying process for this kind of work is borrowed from buying software, when it should be borrowed from hiring a senior person. This guide is the version we'd give a friend: how to read the market, what it costs, what goes wrong and what to ask. At the end, one section explains where Brass Butler sits. It is one of the options, and we say so plainly.
What an AI automation agency actually sells.
“AI agency” covers four quite different businesses. They use the same words on their websites, so the first job is working out which one you are talking to.
Tool resellers
They license a platform, put your logo on it and configure a chatbot, a voice agent or a booking flow from a template. Quick to switch on and cheap to start. You rarely own anything, and when you leave, the system stays with them.
Workflow builders
They connect the software you already have with automation platforms such as Zapier, Make or n8n, often with an AI step in the middle. Good work is common here. The limit is that a workflow does what it was told and nothing more; nobody is watching the business as a whole.
Custom developers
Engineers who build bespoke software around large language models. Capable of anything, priced accordingly, and only as good as the brief you hand them. If you don't know exactly what to build, they will build exactly what you said.
Strategy-first partners
They diagnose before they build, write the plan down, install a system you own and stay on to keep it working as the business changes. Slower to start, because the first step is paid thinking rather than a demo.
None of these is wrong. Each fits a different problem. The trouble starts when a reseller sells you a strategy, or a developer sells you a diagnosis they never did.
What an AI agency charges, and what drives the price.
There is no published rate card for this market, but the quotes owners show us fall into three shapes. The ranges below are the broad market in New Zealand and Australia, not our prices.
By the hour
Freelancers and small shops bill time, from roughly a hundred dollars an hour to well over two hundred and fifty for senior people. Sensible for a bounded fix. Risky for anything open-ended, because nobody is paid to finish.
By the project
A fixed price for a defined build. A simple workflow lands in the low thousands; a custom system wired into your CRM, phones and accounts runs into the tens of thousands. The price is only as good as the scope behind it.
By the month
An ongoing fee for platform licences, hosting, changes and support. Anything from a few hundred dollars a month for a reseller's platform to several thousand for a team actively building. Worth paying when something ships; a tax when nothing does.
Three things move the number more than anything else: how many of your systems the build has to touch, whether a person has to review what the AI sends, and whether the firm has done the same job before. A voice agent that answers the phone and books into one calendar is a small job. A system that reads your inbox, your CRM and your accounts, then drafts the work for your team to approve, is a large one.
Cheap usually hides something. A platform licence you will never own. A template with your logo on it. Scoping you will pay for later in change requests. A junior doing the build while the senior person did the sales call. Or no review layer at all, which is cheap right up until the system sends the wrong email to your best client.
Nine AI agency red flags.
Any one of these is worth a pause. Three together is worth walking away.
- A brief without a diagnosis. They take your description of the problem as the problem. Nobody asks where your hours actually go.
- No ownership of code or data. The system lives in their account, on their platform, and leaves with them.
- Pricing before scoping. A firm quote on the first call means a template, not a build for your business.
- No named person. You meet a salesperson and never learn who builds the thing or who answers when it breaks.
- Demo-ware. The demo is polished, generic and nothing like your business. Ask to see one running in a client's real systems.
- No review layer. Nothing describes what the AI may send, book or pay without a person or a written rule approving it.
- Monthly fees with nothing shipped. The invoice arrives every month; the changes don't.
- No case studies with numbers. Logos and adjectives instead of hours saved, calls answered or revenue moved.
- Tool-first pitches. The conversation starts with the platform they sell, not the work you want off your desk.
Twelve questions to ask on the first call.
These separate the good firms from the ones with good websites. A firm worth hiring will be glad you asked.
- What do you need to know about my business before you can quote?
- Who exactly will build this, and who do I call when it breaks?
- Who owns the code, the data, the prompts and the accounts when you finish?
- What happens to the system if we stop working together?
- Can you show me a client result with a number on it, and can I speak to that client?
- What would you not build for us, and why?
- What can the system send, book or pay without a person seeing it first?
- How do you test it before it goes live with real customers?
- What does the first build include, what does it leave out, and how long will it take?
- What will it cost to run each month once it's live, including platform and usage costs?
- Where is my data stored, and who at your firm can see it?
- What is in the handover pack?
Agency, freelancer, in-house, partner: which suits you.
How to choose an AI agency starts one step earlier: deciding whether an agency is the right shape of help at all. There are four, and each suits a different owner.
An agency
A team that builds to your brief. Good capacity, a range of skills and a process. Strongest when you already know precisely what you want built.
A freelancer
One capable person, usually cheaper and quicker to start. Strongest for a bounded job. The risk is a single point of failure when they move on.
An in-house hire
Someone on your payroll who learns the business properly. Strongest at scale. Expensive, slow to recruit, and hard to judge if nobody else in the firm knows the field.
An AI partner
A firm that diagnoses first, builds a system you own and keeps it working afterwards. Strongest when you know the work is piling up on you but not yet what to build.
AI consulting firms sit somewhere between the first and the last: strong on the diagnosis, sometimes without the build. If you want advice and a plan only, a consultant may be all you need. Our longer view is on the AI consulting page.
What a good first engagement looks like.
Whoever you choose, the first engagement should have four parts. If a proposal is missing one of them, ask why.
- A paid diagnosis. Someone senior spends real hours on your business before anything is quoted. Free diagnoses tend to find exactly what the vendor sells.
- A written plan. What to build, in what order, with a reason and a number behind each step. Yours to keep, whoever builds it.
- Ownership. Code, data, prompts, accounts and documentation belong to you from handover, in writing.
- A bounded first build. One system, a fixed scope, a date measured in weeks, live in your real business. Not a twelve-month roadmap.
The plan should be written for the top brass, the person who signs the cheque and owns the outcome, not for the vendor's sales team. If you can't hand it to another firm and have them build from it, it isn't a plan.
How Brass Butler fits among the options.
Brass Butler is an AI partner for owner-led services businesses in New Zealand and Australia. We are not an agency. We work strategy-first, in the way the last row of the market list describes, and this is how we run the first engagement.
The diagnosis is the AI Strategy Intensive: two hours one-to-one with Titus Mulquiney, our founder, after pre-work that shows where your hours actually go. A written audit, a prioritised plan and a full proposal follow within three business days. It costs $797, it is money-back if it isn't worth it, and it credits in full to a build. The plan is yours to use with us, with another firm or in-house.
The first build is the Foundation: a three-week install of an AI brain, a data layer and a first AI workforce, inside your own systems. It is owned by you from handover, the code, the IP and the documentation. After that, most owners keep us on for AI training, optimisation, implementation and support. We call it kept polished, and the system still belongs to you if you stop. The full method is on how we work.
Proof, with numbers: Ian at AVIT Properties in London went from 80 hours of monthly reporting to two hours of review. Ray White Milldale runs Speed to Lead on the same approach. And we run our own firm on Claude every day. More on the case studies page.
We are not the right choice for everyone. If you already know exactly what you want built, a good workflow builder will do it well, and we will tell you so on the first call.
Broadly, hourly work runs from about a hundred to over two hundred and fifty dollars an hour, project builds from the low thousands to the tens of thousands, and ongoing support from a few hundred to several thousand a month. The scope drives the number far more than the firm does. Get a paid diagnosis before you compare quotes, or you are comparing guesses.
The shapes and the price ranges are much the same. What matters more is whether the firm can work properly at a distance. We build in Auckland and deliver remotely across Australia, and so do plenty of good firms in both directions. Ask how the build, testing and handover run when nobody is in the room.
An agency usually builds to a brief. A consultant usually diagnoses and advises, and may stop there. A partner does the diagnosis, the build and the keeping. Which you need depends on whether you already know what to build. Our AI consulting page covers the consultant side in more depth.
You should, from handover, in writing. That includes the prompts, the automation accounts and the documentation. If a firm can't agree to that, the system is theirs and you are renting it.
A bounded first build should be live in weeks, not months. Ours takes three. Anything quoted over many months as a first step is usually a roadmap dressed up as a project.
Yes, for a single standard job, like a booking flow on a template, where you don't mind not owning it. For anything that touches your clients, your money or your reputation, the cheap quote usually leaves out the scoping, the review layer or the ownership.
A rough picture of where your week goes, the work that keeps landing on you, the software you already pay for, and any quotes you've had. That is enough for a good firm to tell you whether they can help. A firm that doesn't ask for any of it is telling you something.
Choose the firm that diagnoses first.
Whoever you hire, start with a short conversation in which they ask more questions than you do. The discovery call is ours: thirty minutes with Titus on what is landing on your desk, whether a build would take it off, and whether we are the right firm to do it.
Book a discovery callNot ready for a call? Take the three-minute Business Stress Test.