R&D Tax Incentive for AI Development in Australia

The Australian Research and Development Tax Incentive (R&D Tax Incentive) can return a meaningful share of eligible spend to a company building a genuinely new AI product, but most AI builds are not automatically eligible. Whether yours qualifies depends on whether the work is experimental, not on whether it uses AI.

This guide is for founders and small-business owners planning an AI build who want to know how the incentive works, what kind of AI work tends to qualify, and what to organise before you start. It is general information, not tax advice, and you should confirm eligibility with a registered tax adviser.

What is the R&D Tax Incentive and who can use it?

The R&D Tax Incentive is a federal program that gives companies a tax offset for eligible research and development spending. It is administered jointly: the Department of Industry, Science and Resources (through AusIndustry) handles registration of your activities, and the Australian Taxation Office (ATO) handles the tax side.

The size of the offset depends on your turnover. According to the ATO’s rates page, entities with an aggregated turnover of less than $20 million a year that are not controlled by income-tax-exempt entities receive a refundable offset equal to the company tax rate plus an 18.5% premium. Larger entities receive a non-refundable offset with a tiered premium. For an early-stage company, “refundable” is the important word: it can mean cash back even if the company has no tax payable.

Because the figures and rules are updated from time to time, check the ATO page for the income year you are claiming. The business.gov.au program page also notes that the Government announced changes to the incentive in the 2026–27 Budget, starting from 1 July 2028, so planning beyond that date needs fresh advice.

Does building an AI product count as R&D?

Not by default. The test is about the nature of the work. The government’s guidance on core R&D activities says they must be conducted for the purpose of generating new knowledge, and that the outcome cannot be known or determined in advance. It also says the outcome can only be determined by applying a systematic progression of work: forming a hypothesis, running an experiment, observing, evaluating and drawing logical conclusions.

That is a high bar, and it rules out a lot of everyday AI development. Connecting an existing model’s API to your app, wiring up a standard retrieval pipeline or configuring an off-the-shelf tool is usually competent engineering with a predictable result, not experimentation. The question a competent professional would ask is whether the outcome was genuinely unknown before you started.

Which kinds of AI work are more likely to qualify?

Work is more likely to be eligible when you are trying to solve a technical problem where nobody, including a skilled practitioner, can say in advance whether your approach will work. In an AI product, that tends to look like the following situations, and each still needs to be assessed against the legal definition:

• Novel accuracy or reliability problems. You hypothesise that a new combination of retrieval, structured outputs and verification can reach a target error rate on your domain’s data, then design experiments to test it, and the result is genuinely uncertain.
• Unproven architectures for your constraints. You need an approach that works under tight latency, cost or data-residency limits and there is no established pattern to follow, so you test alternatives and record what failed.
• New evaluation methods. Where no accepted way to measure quality exists for your use case, designing and validating one can be an experimental activity in its own right.

By contrast, building the login screen, the billing system, the admin dashboard and the integration with your CRM is almost certainly not core R&D, even though it is essential to the product. Founders who claim everything tend to weaken their position; a tight, well-evidenced claim is more credible than a broad one.

What do you need to do before you start building?

Set up your evidence habits on day one, because the incentive rewards contemporaneous records. The ATO’s checklist for claiming explains that you must register your R&D activities for every income year you want to claim, within 10 months of the end of your company’s income year, and before you claim the offset in your company tax return. It also notes that registration is not a verification of eligibility, so being registered does not protect a weak claim.

In practical terms, that means a short routine:

• Write down the technical hypothesis for each uncertain problem before you attempt it, in plain language.
• Keep a log of experiments: what you tried, what you measured, what happened, and what you concluded.
• Track time and costs against each activity, separating eligible experimental work from routine build work.
• Keep your test sets, evaluation results and failed approaches. Failures are evidence that the outcome was genuinely unknown.
• Agree early with your developer how work will be described and documented, so the records exist without extra effort later.

This is the same discipline that makes AI products better anyway. Our guide to testing AI products with evals describes how to build the test sets that double as R&D evidence.

How does the incentive interact with the way you hire a developer?

Your contract and your developer’s working method both affect how easy a claim is. If you pay an external team to carry out R&D on your behalf, the structure of the arrangement, who owns the results and who bears the technical risk can all matter for who is entitled to claim. This is an area where a tax adviser earns their fee, and you should raise it before you sign, not after the work is done.

It also influences how you scope the project. A staged approach, in which an uncertain technical question is isolated into a short, documented experiment, produces cleaner evidence than one large, fuzzy build. We cover the trade-offs of different contract shapes in fixed price vs time and materials for AI projects, and the broader process of choosing a team in our buyer’s guide to custom AI development in Australia.

What are the common mistakes founders make?

The most frequent mistake is assuming that “AI” equals “R&D”. The second is leaving record-keeping until the end of the financial year, when the details of what was tried and why are already forgotten. The third is missing the registration window, which is a procedural requirement separate from the tax return.

A fourth is treating the incentive as free money in your budget. The offset arrives after you have spent the money, and only if the claim is accepted, so it should not be the thing that makes the project affordable. Build the business case so that it stands on the product’s value, then treat the offset as a possible upside that you confirm with your adviser.

How should you plan an AI build with the incentive in mind?

Start by splitting the product into routine build and genuine technical uncertainty. For the routine part, scope and price it like any software project. For the uncertain part, define the question, the way you will measure success and a limit on how much you will spend finding out.

That approach fits how we run engagements at Neomeric: a free 15-minute scoping call to identify where the real uncertainty sits, then a 2-week Build Sprint at A$6,900 fixed, fully credited toward a pilot, where the riskiest technical question is tested first and documented. Whether any of that work qualifies for the incentive is a question for your tax adviser, but the habit of isolating and recording uncertainty helps either way. For a sense of overall budget, our AI MVP development cost guide sets out what to expect, and the pilot-to-production playbook shows what comes after the first build. If you want to talk it through, book a free scoping call.

Frequently asked questions

What is the R&D Tax Incentive in Australia?

It is a federal program that gives companies a tax offset for eligible research and development spending. AusIndustry handles registration of activities and the ATO handles the tax offset.

How much is the offset for a small AI company?

According to the ATO, entities with aggregated turnover under $20 million that are not controlled by income-tax-exempt entities receive a refundable offset equal to the company tax rate plus an 18.5% premium. Check the ATO rates page for your income year.

Is building an AI app automatically eligible?

No. Core R&D activities must be conducted to generate new knowledge, have an outcome that cannot be known in advance, and follow a systematic progression of work. Routine integration of existing AI services is usually not enough.

When do I have to register R&D activities?

The ATO says you must register for every income year you want to claim, within 10 months of the end of your company’s income year and before claiming the offset in your company tax return. Registration is not a verification of eligibility.

What records should I keep?

Keep your technical hypotheses, experiment logs, test sets, evaluation results, failed approaches, and time and cost records separated by activity.

Should I rely on the offset to fund my AI project?

Treat it as a possible upside rather than a funding source, because it is paid after spending and only if the claim is accepted. Confirm your position with a registered tax adviser.

Neomeric, a Melbourne-based AI product and consulting company — and the team behind NeoMind, Australia’s onshore AI teammates platform — is the local team that scopes and ships AI products every day, and structures builds so the uncertain technical questions are isolated, tested and documented.

Sources

• ATO — Rates of R&D tax incentive offset
• ATO — Checklist for claiming the R&D tax incentive
• business.gov.au — Conducting core R&D activities
• business.gov.au — Research and Development Tax Incentive