How to choose an AI consultant: the questions to ask.

Q: What is the difference between an AI consultant and an AI agency?

A consultant sells a decision: where AI is worth using in your business and what it would take. An agency sells delivery: a tool, an automation or a model, built to a brief you give them. The gap between the two is where most projects fail, because the brief was written before anyone checked which process actually carried the cost. Ask any firm which of the two it is, and who writes the brief.

Q: What should I ask an AI consultant before hiring them?

Eight questions, in the order above: what happens in the first two weeks; who from your business they need; a result in a business like yours with the figure and the person who signs it; what they build on and who owns it afterwards; what they will tell you not to do; who does the work; what it costs and how it ends; and how they test it before your team relies on it. The answers you want are specific, short and about your process, not about the technology.

Q: How much does AI consulting cost in Australia?

There is no market rate, because the work ranges from a two-week diagnostic to a year of building. The useful question is whether the first stage is priced on its own, with a defined end and a decision at the end of it. A firm that can only quote the whole programme is selling the build before anyone knows whether the build is the right answer.

Q: Should we build custom AI or buy software?

Buy when the process is the same as everyone else's: accounting, email, CRM. Build when the standard is yours, the one your best people apply and your competitors cannot copy, because no vendor has it. Most established businesses have two or three of those processes and dozens of the first kind. A good consultant separates them in the first fortnight.

Q: Do we need our own data or IT team first?

No. You need the people who run the process and know what right looks like: the compliance lead, the senior clinician, the site supervisor. The system is built around what they have to prove. Strategy, design and engineering should come from the firm as one team, and the result should work with the systems you already run.

Q: How long should the first engagement be?

Long enough to reach a decision and short enough that the decision is cheap: about two weeks. It should end with the opportunities ranked, a commercial case on each and a clear answer on whether to go on. If the first proposal you receive is for six months, ask what happens in the first two weeks of it.

Most AI engagements are bought the wrong way round: a firm is chosen on its technology, then a brief is written to fit. This guide is for the owner or CEO of an established business who has to pick one, and would rather choose on the questions that predict a result. It names the four kinds of firm you are choosing between, the eight questions to ask each, the answers that matter, and the signs to walk away. We are one of the firms; our own answers are in section five, so you can hold us to them.

In short

Choose on the process, not the technology. The firm that asks which of your processes carries the cost before it names a tool is the one that will find the result.
Eight questions sort them. What happens in the first two weeks, who they need from you, a signed result in a business like yours, what they build on, what they will tell you not to do, who does the work, what it costs and how it ends, how they test it.
The first stage should be short, priced on its own and end in a decision. Anything that starts with a six-month proposal is selling the build before the question is answered.

01Who you are choosing between: four kinds of firm.

The names overlap and every firm claims the whole list, so judge by what each one is paid for.

The AI consultancyPaid for a decision: where AI is worth using, what it would take, whether to proceed. Good when you do not yet know which process to start with. Weak when it hands you a strategy document and leaves the building to someone else.
The automation agencyPaid for tools: workflows, chatbots, integrations between the software you already have. Good for a known, repetitive task with a clear brief. Weak when the brief is wrong, because it will build it anyway.
The software housePaid for a build to a specification: a team, a timeline, a product. Good once the case is proven and the design is clear. Weak at the start, when nobody can write the specification yet.
The studioPaid for the whole arc: find the opportunity, test it, build it, improve it, with strategy, design and engineering in one team. Good when the problem is still a question. Costs more than a tool, because it includes the part where the tool turns out to be the wrong answer.

Ask each firm plainly which of the four it is. The honest ones answer in a sentence.

02Before you call anyone: three things to settle.

What decision you need made. "Should we use AI" is not a decision. "Which of our five processes should we automate first, and is the case strong enough to fund it" is. Write it in one line; it becomes the brief for the first stage.
Who in your business holds the standard. Every process worth automating has one or two people who know what right looks like. A firm will need them for a few hours a week in the first stage. If you cannot free them, the engagement will stall whoever you hire. The key person risk guide shows how to find them.
What a result would be worth. Not a model of the business case: the cost of the mistake, the hours of checking, the work you turn away. The firm that cannot turn those into a number in two weeks will not turn them into a result in six months.

03The questions to ask, and the answers that matter.

Ask every shortlisted firm the same eight. Write the answers down, because the differences are in the wording.

The eight questions

Ask each firm the same eight, in this order.

What happens in the first two weeks? A good answer names the people they will sit with, the processes they will map and the decision you will have at the end. A bad answer is a workshop and a roadmap.
Who from my business do you need, and for how long? A good answer names roles, the compliance lead, the senior clinician, the supervisor, and asks for hours, not days. A bad answer is "a project sponsor".
Show me a result in a business like mine. A good answer has a figure, a before and after, and the name of the person in the client's business who signs it. A bad answer is a logo wall.
What do you build on, and who owns it afterwards? A good answer names the platforms, says what you own, and says what it costs to run without them. A bad answer is a product they resell.
What will you tell me not to do? A good answer includes at least one thing you asked for. A firm that agrees with everything is selling, not advising.
Who does the work, and who do I talk to? A good answer puts the people who sold it in the room through delivery. A bad answer is a senior pitch and a junior team.
What does it cost, and how does it end? A good answer prices the first stage on its own and tells you what you leave with if you stop there. A bad answer only quotes the whole programme.
How do you test it before my team relies on it? A good answer describes the test in your conditions: real consignments, real reports, a real site. A bad answer is a demo.

04Red flags, in the order you will meet them.

A proposal before a question. If the scope arrives before anyone has asked which process carries the cost, the scope is a guess.
A demo of the model, not of your process. The technology is not the hard part. The hard part is the standard your best people apply, and whether the system can hold it.
"AI strategy" with no process named. A strategy that does not say which job changes first, for whom, by how much, is a document.
The senior team sells, a different team delivers. Ask who will be in the room in week six.
A platform they resell. The recommendation will always be the platform.
No test in the field. A system nobody has run on real jobs, in real conditions, before your team relies on it, is a prototype whatever it is called.
Time measured in quarters before a result. The first result should be visible inside the first stage, even if it is only a ranked list and a number.

05How we answer them, so you can hold us to it.

We are the fourth kind of firm: one team of strategy, design and engineering, paid for the arc from question to running system. Our answers to the eight, in order.

The first two weeks are spent with the people who run the process, mapping it against real jobs. They end with the opportunities ranked, a commercial case on each and a decision on whether to go on.
From you we need the people who hold the standard, for a few hours a week, and someone who can make the decision at the end.
Results in businesses like yours: compliance paperwork per run from 8 hours to under 30 minutes at Xtreme Freight; a clinical report from around 4 hours to about 20 minutes at Virtuosa; an emergency alert to the right people in under ten seconds on a mine site. Each case study names the person who signs the result.
We build on low-code foundations where the rules will change and custom engineering where they will not, and we say which before we start. You own the system and the data.
We will tell you not to automate a process nobody has standardised, and not to start with the one that is easiest rather than the one that costs most.
The partners who run the first stage stay through design, build and the stage after launch.
Cost and ending: the first stage is priced on its own and ends in a decision. A prototype is one sprint; larger design and build work typically runs across 5 to 12 sprints. If you stop after the first stage you leave with the ranked list and the cases.
Testing happens in your conditions before anyone relies on the system: real consignments at Xtreme Freight, a psychologist signing a drafted report at Virtuosa, field conditions on site at AtSite.

<30minDGC with Xtreme Freight. Compliance paperwork per run, down from 8 hours. Zero paperwork errors across 120 tonnes in the first pilot.Read the case study
20minVirtuosa AI. Per clinical report, down from around 4 hours. Clinics see 25% more clients per psychologist.Read the case study
<10sAtSite. Emergency alerts to the right people on a mine site, with privacy-first geofencing and compliance reporting from the same record.Read the case study

We help established businesses identify where AI can create commercial value, then bring strategy, design and engineering together to deliver it. The process is the one described in the answers above.

06Questions buyers ask before the call.

What is the difference between an AI consultant and an AI agency?

A consultant sells a decision: where AI is worth using in your business and what it would take. An agency sells delivery: a tool, an automation or a model, built to a brief you give them. The gap between the two is where most projects fail, because the brief was written before anyone checked which process actually carried the cost. Ask any firm which of the two it is, and who writes the brief.

What should I ask an AI consultant before hiring them?

Eight questions, in the order above: what happens in the first two weeks; who from your business they need; a result in a business like yours with the figure and the person who signs it; what they build on and who owns it afterwards; what they will tell you not to do; who does the work; what it costs and how it ends; and how they test it before your team relies on it. The answers you want are specific, short and about your process, not about the technology.

How much does AI consulting cost in Australia?

There is no market rate, because the work ranges from a two-week diagnostic to a year of building. The useful question is whether the first stage is priced on its own, with a defined end and a decision at the end of it. A firm that can only quote the whole programme is selling the build before anyone knows whether the build is the right answer.

Should we build custom AI or buy software?

Buy when the process is the same as everyone else's: accounting, email, CRM. Build when the standard is yours, the one your best people apply and your competitors cannot copy, because no vendor has it. Most established businesses have two or three of those processes and dozens of the first kind. A good consultant separates them in the first fortnight.

Do we need our own data or IT team first?

No. You need the people who run the process and know what right looks like: the compliance lead, the senior clinician, the site supervisor. The system is built around what they have to prove. Strategy, design and engineering should come from the firm as one team, and the result should work with the systems you already run.

How long should the first engagement be?

Long enough to reach a decision and short enough that the decision is cheap: about two weeks. It should end with the opportunities ranked, a commercial case on each and a clear answer on whether to go on. If the first proposal you receive is for six months, ask what happens in the first two weeks of it.

Author

Gabriela Diaz Reyes, Mayfly Ventures, Melbourne. Written and reviewed 5 November 2026. Corrections and questions are welcome.

Keep reading

Ask us the eight questions first.

A free 30 minute call. Bring the process you have in mind or the proposal you have been sent; we will answer the eight and tell you whether the first stage is worth running. No technical brief needed.

Book a discovery call See how we work →

Free 30 minute call · Melbourne, Australia