How to Choose an AI Development Company in Australia: 12 Questions to Ask

The short answer

Pick the AI development company that can show you a system it built running in production, will keep your customers’ personal information in an Australian cloud region, and puts in writing that you own the code. At Zeebrix, a first AI project costs from about A$15,000 for one AI feature added to an existing product to about A$105,000 for a full AI MVP, quoted as a fixed price.

Before writing this we read the guides that rank for this search. Most are written by Australian agencies and repeat sensible advice: check the track record, ask about support, don’t buy on price alone. None of the ones we read mentioned the Privacy Act, GST, the R&D Tax Incentive or who owns the code by default under Australian law. Those are the places Australian buyers get caught.

Here are the 12 questions we would ask in your position.

What is an AI development company, and what are you buying?

An AI development company is a firm that designs, builds and maintains software in which a language model or machine learning model does part of the work, and that takes responsibility for how the software behaves once real people use it.

That last clause is what separates a development company from someone wiring a chatbot to an API. A demo can be made to work over a weekend. Getting the right answer to the four hundredth odd question from a customer, without leaking anything it shouldn’t, is engineering.

So when you compare firms you are buying three things. The build. The testing that proves it works on your data. And someone to call when it misbehaves in month seven. Price only the first one and the other two will disappoint you.

Questions 1 to 3: Have they shipped this before, and how will they prove it works?

Start here. These three questions filter out most of the field in a single meeting.

Our own case studies include Mindway, a mental wellness app, and Epic Water, an IoT water management app.

• 1. Can you show me an AI system you built that real customers or staff have been using for months? Ask to see it running, not a slide about it. A prototype only proves they can build a prototype
• 2. How will we know it is right? A serious firm proposes a test set of real questions or documents from your business, agrees it with you before the build, and reports results against it every week. “We’ll tune it until you’re happy” gives you nothing to hold them to
• 3. What do you need from our data, and what happens if it is messier than you expect? The honest answer involves a short look at the data before any fixed price. A firm that prices a document assistant without opening the documents is guessing

Questions 4 to 6: Where will your data live, and does the Privacy Act apply?

This is where an Australian checklist parts ways with an American one. The Office of the Australian Information Commissioner (OAIC) says the Privacy Act 1988 covers organisations with annual turnover above A$3 million, plus some smaller ones, including private health service providers and businesses that buy or sell personal information. If you are covered, your AI system has to meet the 13 Australian Privacy Principles.

Two of the principles do most of the work here. APP 8 sets out the steps you must take before personal information is disclosed overseas, and sending a customer’s message to a model hosted in the United States can count. APP 11 requires reasonable steps to protect the information you hold. The OAIC’s Guidance on privacy and the use of commercially available AI products goes further and recommends, as best practice, that organisations don’t enter personal information into publicly available generative AI tools at all.

There is a new date to plan around too. From 10 December 2026, under changes made by the Privacy and Other Legislation Amendment Act 2024, your privacy policy must explain where computer programs make decisions, or do a step closely tied to decisions, that significantly affect people. The OAIC published guidance on it on 30 September 2026. If the system you are buying approves, declines, ranks or flags people, raise this in the first meeting.

• 4. Which cloud region will our data and our model calls run in? For most Australian projects the answer should be AWS Sydney or Melbourne, or Azure Australia East, with the model called in the same region where it is offered
• 5. Does any personal information leave Australia, even for logging or debugging? Ask for a list of every third-party service in the design. Analytics and error-tracking tools are the usual leaks
• 6. Will our data be used to train anyone’s model? The answer should be no, in writing, from both the agency and the model provider

Questions 7 and 8: Who owns the code, and who actually writes it?

Most buyers assume that paying for software means owning it. Under Australian law that isn’t the default. IP Australia states that IP created by a contractor belongs to the contractor unless the contract says otherwise, and business.gov.au repeats the warning in its guidance on hiring contractors. Your contract needs a clause assigning the code, prompts, test sets and any fine-tuned models to you.

The second question is about people. Plenty of agencies that look local run engineering offshore. That can work well. What doesn’t work is finding out after you sign.

We run a mixed model and say so up front. Our Perth office is where Australian clients meet us and where projects are managed, and engineering is shared with our team in San Jose, which gives your project a longer working day.

• 7. Will the contract assign all IP to us, and will the code, cloud accounts and model API keys sit in accounts we own from day one? If the agency hosts everything under its own login, leaving later turns into a negotiation
• 8. Who exactly will write the code, where are they, and can I meet them before I sign? Ask how many working hours they overlap with yours, and who picks up the phone when production breaks at 9 am in Perth

Questions 9 and 10: Fixed price or hourly, and is GST on top?

Hourly billing is common in AI work because outcomes are uncertain. The uncertainty is real. Hourly billing simply hands all of it to you. A fixed price for a fixed scope, set after a short discovery phase, puts the cost of a wrong estimate on the people who made the estimate.

The trade-off is flexibility. Under a fixed price, a new idea halfway through becomes a change request with its own price. We think that is healthy, because every addition becomes a decision someone made on purpose.

Our own prices are converted from US dollars at about A$1.50 to US$1, so the AUD figures are approximate. An AI feature added to an existing product is about A$15,000 to A$37,500. A custom AI app or chatbot on your own data is about A$22,500 to A$60,000 over 4 to 8 weeks. An AI agent connected to two or three of your systems is about A$27,000 to A$60,000, and an AI MVP for a startup is about A$45,000 to A$105,000 over 6 to 10 weeks. The full breakdown is in our guide to AI development cost in Australia.

• 9. Is this a fixed price, what exactly is in scope, and how are changes priced? Ask too what hosting and model usage will cost each month after launch
• 10. Is GST included, added on top, or not charged at all? A GST-registered Australian supplier adds 10%. An overseas supplier usually won’t charge GST to a registered business that quotes its ABN, but the ATO’s reverse charge rules can still require you to account for the GST yourself if you can’t claim a full credit. Ask your accountant

Question 11: Could part of the build qualify for the R&D Tax Incentive?

If your company’s aggregated turnover is under A$20 million, eligible R&D spending earns a refundable tax offset equal to your company tax rate plus an 18.5% premium, which is 43.5% for most small companies. Because it is refundable, a company that isn’t yet profitable can receive it as cash. The business.gov.au site also sets the deadline: you register within 10 months of the end of the income year in which the work happened.

Now the caution. Only genuine experimental work qualifies as core R&D, meaning systematic, hypothesis-driven activity whose outcome could not be known in advance. Connecting a chatbot to your help desk does not qualify. Testing whether a new retrieval method can answer questions from scanned engineering drawings at an acceptable error rate might. The same site publishes a guide called Software activities and the R&D Tax Incentive, and an R&D tax adviser should see your plan before work starts. The program also generally requires at least A$20,000 of R&D spending in the year.

• 11. Will you keep records of hypotheses, experiments and results in a form my R&D tax adviser can use? A firm that doesn’t know what you mean has probably never had a client make a claim

Question 12: Do they follow the Government’s Guidance for AI Adoption?

The Australian Government’s main AI guidance for businesses is voluntary. In October 2025 the National AI Centre, part of the Department of Industry, Science and Resources, published the Guidance for AI Adoption on industry.gov.au. It updates the 2024 Voluntary AI Safety Standard and condenses its 10 guardrails into six practices: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, and maintain human control.

You don’t need a vendor with a certificate. You do want one who has read it, because some of those practices land on the supplier directly. The guidance tells businesses to get proof of testing from their suppliers and to keep people able to override an AI system.

• 12. Which parts of the Guidance for AI Adoption will you help us meet, and what will you hand over? Good answers include an entry for your AI register, the test results, and a note of where a person can override the system

When is a freelancer or a big consultancy the better choice?

We are a mid-sized agency, so weigh this accordingly. An agency like ours isn’t always the right buy.

Hire a local freelancer when the job is small and tightly defined (say, one automation inside a tool you already use) and you have someone in-house who can review the work. You will pay less, and an agency’s project management adds little to a job that size.

Hire a large consultancy when you are a listed company or a government body with formal procurement rules, when the AI work is one part of a multi-year transformation, or when you need the vendor to sit on an approved panel and carry large insurance cover. You will pay more and get more process. Sometimes that process is exactly what your board needs to see.

An agency fits between the two: projects that need AI, backend, mobile and integration work at once, with one team accountable for a fixed price. If you can’t tell which group you are in, a short AI consulting engagement with a working prototype, about A$12,000 to A$27,000 with us, will settle it before you commit to a build.

Your checklist before you sign

Take this into your final meeting with each shortlisted firm, and write down any answer that is vague.

• A live production system shown to you on screen
• An agreed test set and accuracy target before the build starts
• Data and model calls hosted in an Australian region, with every overseas service listed
• A written promise that your data won’t train anyone’s model
• A privacy review against the APPs, plus a plan for the 10 December 2026 automated decision rules if they apply to you
• Full IP assignment, with code and cloud accounts in your name from day one
• Named engineers you have spoken to, and agreed hours of overlap
• A fixed price, a written scope and a change-request process
• GST treatment confirmed in writing
• R&D records kept from week one if you plan to claim
• A support plan with a monthly cost and agreed response times

Frequently asked questions

Related reading