---
title: How to choose an AI consultant or developer: twelve questions to ask
url: https://generalmatters.ai/briefings/how-to-choose-an-ai-consultant/
date: 2026-10-08
updated: 2026-10-08
author: General Matters AI
---

# How to choose an AI consultant or developer: twelve questions to ask

Twelve plain questions that separate a useful AI consultant from a sales pitch, with what a good and a bad answer sounds like. For Australian small business.

## In short

- Choose the person who starts with your problem, shows you something live, tells you what they will not do, puts ownership in writing, and names who looks after it afterwards.
- Write the problem in plain words before you talk to anyone. A good consultant will ask better questions than a vendor will answer.
- Twelve questions follow, grouped into the work, ownership, afterwards and the person, each with what a good and a bad answer sounds like.
- Run a small, paid, time limited trial on one real job before any larger commitment.
- You stay accountable for how AI is used in your business, so choose someone who helps you keep control.

## The short answer
Choose the AI consultant or developer who starts with your problem rather than their product, can show you something live that real people use, tells you plainly what they will not do, puts ownership of the work in writing, and names who looks after it after launch. Everything else is detail. This guide gives you twelve questions that test those five things, and what a good and a bad answer sounds like.

There is a lot of AI for sale, and much of the selling aims to make you feel behind. You do not need to be an expert to buy well. You need a clear problem and a few good questions.

## Before you talk to anyone
Write the problem down in plain words. Not "we want to use AI", but something like "invoice entry takes our office manager about ten hours a week" or "enquiries arrive by email, phone and a form, and some get missed". Add how often it happens, who is affected and what it costs you.

This does two things. It stops the first conversation turning into a demonstration of someone else's product. And it gives you a way to judge every answer you hear: does this person start from the problem, or from the tool? If you are not sure you are ready, our [readiness checklist](/briefings/ai-readiness-checklist-small-business/) will help you work that out first.

## Twelve questions to ask
### About the work
#### 1. What exactly will it do on a Tuesday?
Ask them to describe a normal day with the thing running, not a list of capabilities. **A good answer** is specific: who uses it, what they see and what changes. **A warning sign** is a long answer about transformation that never mentions a person doing a task.

#### 2. How will we know it works?
Ask how success will be judged, in your terms, using your real examples. **A good answer** names a measure you already care about, such as hours saved, errors avoided or enquiries answered, and agrees it before work starts. **A warning sign** is "you will just see the difference".

#### 3. What will you not build?
Ask what is out of scope and what they would advise against. **A good answer** names things, including parts of your own plan that are not worth doing. **A warning sign** is a consultant who can do everything, which usually means they are guessing.

#### 4. What happens when it is wrong?
Every system is wrong sometimes. Ask how errors are caught, who sees them and who is responsible. **A good answer** describes checks, a person in the loop and a way to switch it off. **A warning sign** is confidence that it will not happen.

### About ownership
#### 5. Who owns the code, the data and the accounts?
The answer should be you, in writing. **A good answer** is a clear yes, with accounts opened in your business name. **A warning sign** is "you have access to our platform". Our [ownership guide](/briefings/who-owns-your-website-or-software/) covers what to look for.

#### 6. Can another firm take over?
Ask what another builder would need to pick it up. **A good answer** lists the code, the documentation and the accounts, and agrees to help with a handover. **A warning sign** is silence, or a reason it cannot be done.

#### 7. Where does our information go?
Ask what the system reads, where it is stored, who can see it and whether it is used to train anyone's models. **A good answer** is specific and offers it in writing. **A warning sign** is vagueness, or an assurance with no detail.

### About afterwards
#### 8. Who looks after it after launch?
Most of the cost and most of the trouble come later. Ask who monitors it, who fixes it, how fast and on what terms. **A good answer** names a person and a process. **A warning sign** is that this part will be sorted out later.

#### 9. What if you are unavailable?
One person who holds everything in their head is a risk. Ask what is written down and who else could help. **A good answer** points to documentation and a second person. **A warning sign** is "that will not be an issue".

#### 10. What happens if we part ways?
A good answer is calm and specific: what you take, how long it takes and what help they give. **A warning sign** is a defensive answer, or one that starts with "that will not be necessary".

### About the person
#### 11. Can I speak to someone you have worked with?
A short call with a past client tells you more than a portfolio. **A good answer** is a name and a number, and a client who describes the work in plain terms. **A warning sign** is only testimonials, or clients who cannot be reached.

#### 12. Who will actually do the work?
Be clear whether it is the person you are meeting or someone you will never speak to. **A good answer** names the people and how you will communicate. **A warning sign** is a team that appears after you sign.

*Diagram: Four groups of questions. The work: what does it do on a Tuesday, how will we know it works, what will you not build, what happens when it is wrong. Ownership: who owns the code, data and accounts, can another firm take over, where does our information go. Afterwards: who looks after it, what if you are unavailable, what happens if we part ways. The person: can I speak to a past client, who will actually do the work.*

## Red flags in one list
- Urgency: a deadline, a limited offer, a fear that competitors are ahead.
- Vocabulary you did not ask for and cannot check.
- A demonstration on invented data that never touches your own.
- No mention of anything the thing will not do.
- A price that depends on how impressed you are in the room.
- A platform you must keep paying to keep using.

None of these means the person is dishonest. Together they mean you should slow down.

## Test before you commit
Instead of a roadmap, ask for one small, real job on your own information, with a way to judge it and a fixed time limit. Two to four weeks is usually enough. Run it beside your normal process, and have someone check every output.

You learn how they work, how they communicate and whether the thing is useful, at low cost and low risk. Agree beforehand what you will have at the end, including the documentation and accounts, so that the trial is something you keep whether or not you continue.

*Diagram: A four step flow followed by a decision. Write the problem in plain words. Agree what success looks like. Run a two to four week trial beside your normal process. Compare, with a person checking every output. Then keep, change or stop. The trial should be something you keep, with its documentation and accounts, whether or not you continue.*

## You stay accountable
Whoever you choose, you remain responsible for how AI is used in your business. The Australian Government's Guidance for AI Adoption puts accountability first among its six essential practices. In practice, that means a named person on your side who understands what the system does, checks its output and can switch it off. A good consultant helps you keep that control. A poor one makes you dependent on them for it.

*Image: Washerwomen at the Edge of the Pond. Eugène Boudin.*

## Consultant, freelancer, agency or do it yourself?
| Option | Good for | Watch for |
|---|---|---|
| A specialist consultant or small firm | A specific problem, close contact, handover | Dependence on one or two people |
| A freelancer | A small, well defined build | Availability, and who looks after it later |
| A large agency | Large, complex or regulated work | Layers, handoffs and a lead who is not the doer |
| Doing it yourself with off the shelf tools | Personal productivity, low risk tasks | Time cost, inconsistent results and customer data in tools you have not checked |

## Frequently asked questions
### How do I choose an AI consultant?
Start from your problem, then ask twelve questions covering the work, ownership, what happens afterwards and the person. Prefer someone who shows you something live, tells you what they will not build and puts ownership in writing.

### Do I need an AI consultant or a developer?
If you are not sure what to build or whether AI is the answer, you want advice first. If you know the job and want it built, you want a developer. Many good small firms do both, and the best ones will tell you when you need neither.

### How much does an AI consultant cost?
It varies widely and depends on scope, so ask for a fixed outcome on one real job before you commit. Be wary of a single flat number with no breakdown, and of pricing that depends on how impressed you are.

### How long should a first project take?
A well defined first job can be tried in a few weeks. A fixed time limit and a clear definition of success matter more than the exact length.

### Should the consultant be local?
Not necessarily, but being able to talk to a real person, in your time zone, who understands how Australian businesses work is useful. It matters most for the discovery and handover stages.

### What if I already tried an AI tool and it did not work?
That is common and usually not the tool's fault. It was probably a general tool used on a general task with no checks. A specific job, your own information and a named owner give a different result.

## References

- [Guidance for AI adoption (Department of Industry, Science and Resources)](https://www.industry.gov.au/publications/guidance-for-ai-adoption)
- [Does My Business Own the Software it is Having Developed? (LegalVision)](https://legalvision.com.au/own-the-software/)