The R&D Tax Incentive and AI: Eligibility, Claim Rates and What Australian Businesses Get Wrong product guide

I now have comprehensive, authoritative data to write the article. Let me compose the final, verified piece.

The R&D Tax Incentive and AI: Eligibility, Claim Rates and What Australian Businesses Get Wrong

Australia's R&D Tax Incentive (R&DTI) is the federal government's single largest lever for stimulating private-sector innovation — and it is increasingly being used to fund artificial intelligence and machine learning development. Yet it is also one of the most widely misunderstood programs in the Australian business support landscape. Businesses routinely over-claim, misclassify routine software work as experimental research, and submit documentation that would not survive a compliance review. At a time when the government is actively signalling its commitment to AI-led economic growth — projecting contributions of up to $600 billion annually to GDP by 2030 — getting the R&DTI right for AI projects has never been more consequential.

This guide cuts through the complexity. It explains the precise legislative test that applies, the financial value at stake, the specific errors that AI and machine learning companies make most often, and what documentation standard is required to withstand an ATO or DISR audit. For the broader context of how the R&DTI fits within Australia's full suite of AI funding mechanisms, see our guide to Every Australian Government AI Grant and Funding Program.

The Scale of AI-Related R&DTI Claims in Australia

The financial stakes are significant and growing. Around A$950 million has been registered by businesses for activities associated with AI under the R&D Tax Incentive program, across the 2022–23 and 2023–24 income years. This figure, published by Austrade, reflects the rapid acceleration of AI investment in Australia's business sector.

The underlying R&D expenditure data from the ABS confirms the trend. Businesses more than doubled their investment, putting $668.3 million into AI R&D in 2023–24, compared to $276.3 million in 2021–22 — showing how rapidly Australian firms have begun investing in AI. Specifically, the strongest growth in R&D was seen in the Information and computing sciences research field, which includes spending on Artificial Intelligence — a category that grew by 142 per cent since 2021–22.

The R&D tax incentive program is the Australian Government's most significant lever for funding innovation and R&D, playing a pivotal role in shaping the nation's economic future and bolstering Australia's global competitiveness. Against this backdrop, it is all the more important that AI businesses claim correctly — not just to protect themselves from audit risk, but to preserve the program's integrity for the broader innovation ecosystem. (For how AI R&D investment data contextualises Australia's broader adoption patterns, see our article on Australian AI Adoption Statistics.)

The Legislative Test: What the Law Actually Says

The Definition of Core R&D Activity Under Div 355-25

The R&DTI does not have a special category for AI or machine learning. The R&D tax incentive legislation requires businesses to meet the definition of an R&D core activity to be eligible. This is defined in Div 355-25 of the Income Tax Assessment Act as activities whose outcome cannot be known or determined in advance on the basis of current knowledge, information or experience, but can only be determined by applying a systematic progression of work based on principles of established science. This is a generic eligibility requirement and makes no special mention of software or AI.

The full statutory definition, as set out in section 355-25(1) of the ITAA 1997, requires that core R&D activities are experimental activities that:

(a) whose outcome cannot be known or determined in advance on the basis of current knowledge, information or experience, but can only be determined by applying a systematic progression of work that: (i) is based on principles of established science; and (ii) proceeds from hypothesis to experiment, observation and evaluation, and leads to logical conclusions; and (b) that are conducted for the purpose of generating new knowledge (including new knowledge in the form of new or improved materials, products, devices, processes or services).

Critically, the knowledge gap must be assessed against what is known globally, not just within your organisation. Current knowledge, information or experience refers to knowledge that is reasonably accessible on a worldwide basis — it is knowledge available at the time you start your R&D activities — and means the facts, information and skills acquired through experience or education, and the theoretical or practical understanding of a subject.

The object of the entire Division, as stated in section 355-5(1) of the ITAA 1997, is to encourage industry to conduct research and development activities that might otherwise not be conducted because of an uncertain return from the activities, in cases where the knowledge gained is likely to benefit the wider Australian economy.

Core vs Supporting R&D Activities

Division 355 of the ITAA 1997 provides taxpayers with access to certain tax concessions in respect of R&D activities they carry out. R&D activities are defined to include both core R&D activities and supporting R&D activities. Supporting activities, defined under section 355-30, are activities that are undertaken for the dominant purpose of supporting core R&D activities. They do not need to be experimental in themselves, but they must have a direct and necessary relationship to the core activity.

What the R&DTI Is Worth for AI Companies

The R&D Tax Incentive offers a refundable tax offset of up to 43.5% for eligible R&D activities, for company groups with aggregated turnover less than $20 million.

For AI and machine learning, which often require significant computing resources and expensive hardware, this refund can significantly improve cash flow for start-ups.

For larger companies — those with aggregated annual turnover of $20 million or more — a non-refundable tax offset applies at a lower rate. Businesses must spend at least $20,000 on eligible R&D activities within the financial year to access the program, with an exception where expenses are incurred through a registered Research Service Provider (RSP) or a Cooperative Research Centre (CRC).

Approximately 12,000 businesses claim the R&D Tax Incentive each year. 48% of businesses claiming are small businesses with fewer than 20 employees. 43% of all claims are from the professional, scientific and technical services industry. This concentration in professional and technical services means that AI and software companies are already the program's heaviest users — and, as a consequence, its most scrutinised.

What Qualifies: AI Activities That Can Meet the Test

AI development is inherently experimental. Training a machine learning model doesn't come with a guaranteed outcome — you form a hypothesis, design experiments, iterate on architectures, and test results against benchmarks. That process of systematic investigation and refinement is exactly what the R&DTI was built to reward. Unlike off-the-shelf software implementation, genuine AI development involves unknown outcomes at the outset.

Activities like developing novel algorithms, training and evaluating ML models, building proprietary AI infrastructure, and researching the application of AI to domain-specific problems all have strong alignment with the R&DTI's core eligibility criteria.

Specific examples that are likely to qualify (when properly documented) include:

• Developing novel machine learning architectures for specific use cases; researching and testing approaches to training data curation and augmentation; designing proprietary natural language processing (NLP) or computer vision pipelines; investigating reinforcement learning approaches for complex decision-making problems; and building and evaluating AI models where performance thresholds are not guaranteed.

• Previously, AI has been included in R&D tax projects that involved predictive algorithms, machine learning and Natural Language Programming for human-machine interfaces.

The decisive question is always the same: the key phrase that trips up many claimants is "unknown outcome." If a competent AI engineer could reliably predict the result using existing knowledge and techniques, the activity may not qualify as core R&D. But if you're pushing into territory where standard methods fail or haven't been applied to your specific domain, that's where genuine core R&D lives.

The Most Common Errors Australian AI Businesses Make

Error 1: Treating All AI Development as Automatically Eligible

With the recent buzz in artificial intelligence (AI) and machine learning (ML), companies must be careful not to assume that activities around AI are automatically eligible under the R&DTI. This is the most pervasive mistake. The use of AI tools, frameworks, or techniques does not, by itself, constitute eligible R&D.

The inclusion of AI or ML techniques does not, in and of itself, constitute eligibility for the R&D Tax Incentive. The requirements for activities to be conducted for the purpose of generating new knowledge, and to have uncertain scientific outcomes, should be referenced to the specific field of inquiry.

Error 2: Claiming Routine Software Development as Core R&D

The legislative definition of 'eligible R&D' for the R&D Tax Incentive is different from the usual business definition of R&D. In practice, R&D claimants who self-assess their eligibility often incorrectly classify complex but 'routine' software development as 'experimental'.

Eligible core R&D is not learning how to use existing products, technologies or techniques in the manner in which they are designed to be used. Eligible R&D is not using such products, technologies or techniques in a specific application. Software development activities can pose a challenge to the self-assessment of eligible activities because the process of developing, modifying or customising software is superficially similar to the eligibility requirements for core R&D activities. They are by definition systematic and can be iterative and cyclical, and almost always involve testing. However, they are not necessarily experimental as required under the programme's legislation.

Error 3: Applying Known Processes to New Business Parameters

Where we see scope for potential pitfalls in AI claims is where companies may be applying known processes — including established software development, AI, ML and data science processes — to their commercial parameters and workflows in a standard or known manner. For example, a company that customises or adapts existing or known AI technology to automate their business operations without generating any new technical knowledge or capabilities would be unlikely to be conducting R&D activity.

If companies apply known processes, such as known software development, AI and ML processes, to their parameters, they are unlikely to qualify for the R&D Tax Incentive. In this scenario, the company is not generating new knowledge in the form of a new product or process — rather, they are applying existing knowledge to their business operations.

Error 4: Treating Prompt Engineering as R&D

This is an emerging error category that has not yet attracted specific ATO guidance but follows directly from the legislative test. Prompt engineering — the practice of crafting inputs to optimise outputs from a pre-existing large language model — applies a known technique to a known model. The outcome may be uncertain in a commercial sense, but it does not constitute a scientific or technical uncertainty that can only be resolved through systematic experimentation. Businesses that classify prompt engineering workflows as core R&D activities are particularly exposed in a compliance review.

Error 5: Misidentifying the ANZSRC Research Code

The R&D Tax Incentive application requires businesses to specify the relevant Australian and New Zealand Standard Research Classification (ANZSRC) code. While many AI/ML founders might assume "Artificial Intelligence" or "Machine Learning" (4611) is the appropriate code, its scope is limited.

This distinction is critical because the R&D Tax Incentive excludes certain research areas, including social sciences, arts, humanities, and market research into consumer behaviour and preferences. Consequently, activities applying machine learning to answer research questions in these areas may be ineligible for registration.

Documentation: What You Need to Survive an Audit

The R&DTI program is a self-assessment regime. Receiving a registration number from DISR doesn't mean the R&D activities meet the eligibility requirements. Both the ATO and DISR conduct independent compliance reviews, and the documentation standard is demanding.

The ATO and the Department of Industry, Science and Resources (DISR) jointly administer the R&D tax incentive. Your R&D activities must be registered with DISR before claiming the tax offset. But registration is not a safe harbour. Registering your activities doesn't mean you'll automatically receive the offset at tax time — the ATO conducts its own reviews of claims being made. It's very important to remember that the ATO and DISR both conduct compliance reviews and audits of claims to make sure the law is being followed.

The ATO has increased its audits of R&DTI claims, especially those leading to large refunds. Refunds are often delayed while taxpayers complete detailed questionnaires to explain and justify the methods used to calculate each category of R&D expenditure. If a claim is rejected, penalties can be steep — the ATO generally seeks to impose a penalty equal to 50% of the tax shortfall, citing recklessness.

DISR's Documentation Expectations for AI/ML Claims

DISR's guidance highlights its expectations that you: compile contemporaneous documentation before commencing any experimentation; work to and can document a specific hypothesis or hypotheses; and can describe how you intend to validate a technical or scientific challenge.

The DISR's published Hypothetical Machine Learning case study — which focuses on a company developing an irrigation decision support system incorporating satellite imagery — provides a practical benchmark. The case study suggests a suitable process to determine the lack of available knowledge: a preliminary literature review to identify unanswered research questions; hypothesis development based on identified gaps; and a secondary literature review to find any follow-up research testing the proposed solution. If the second review finds no research testing the solution, the business can conclude that the activity's outcome was unknown based on current knowledge.

The Documentation Framework: What to Prepare

The following documentation framework reflects the combined expectations of DISR and the ATO for AI and machine learning R&D claims:

Maintaining comprehensive documentation throughout your R&D program is non-negotiable — both DISR and the ATO place heavy emphasis on documentation when reviewing claims. Real-time evidence not only demonstrates your self-assessment process but also directly aligns with the program requirements.

In June 2024, DISR released updated versions of key guidance materials — the 'R&D Tax Incentive Guide to Interpretation' and 'Software-related activities and the R&D Tax Incentive'. The updated Guide to Interpretation provides enhanced detail on record-keeping expectations and clarification on what constitutes 'current knowledge'. Businesses should review these updated documents before lodging any AI-related claim.

The New Compliance Environment

From 15 August 2025, a redesigned online R&D application form has been used, offering clearer guidance to align activities with statutory requirements.

From 1 July 2025, DISR has implemented a One-Strike Policy: if one eligibility requirement is not met, assessors may issue a decision without reviewing the remainder of your claim. Every eligibility box must be ticked before submission.

Businesses that receive government grants alongside their R&DTI claims also need to be aware of an important interaction. The ATO now states that if you receive a government grant or similar funding, you should apply the "expenditure not at risk" rule before clawback. This differs from the prior understanding that grants were clawback-only. The change could reduce claimable amounts for some grant-funded R&D projects. This is especially relevant for businesses accessing the AI Adopt Program or other co-funded initiatives alongside an R&DTI claim (see our guide on How to Apply for Australian Government AI Grants).

Key Takeaways

• Around A$950 million has been registered by businesses for activities associated with AI under the R&D Tax Incentive program across the 2022–23 and 2023–24 income years — demonstrating both the scale of AI R&D in Australia and the program's importance to the sector.

• Core R&D activities are defined under s.355-25 of the ITAA 1997 as experimental activities whose outcome cannot be known or determined in advance, that proceed from hypothesis to experiment and evaluation, and are conducted for the purpose of generating new knowledge. AI projects must satisfy this test on the technical merits — the use of AI tools alone is not sufficient.

• Companies often self-assess all of their ML/AI software development as R&D when some aspects may be routine in nature. If companies apply known processes, such as known software development, AI and ML processes, to their parameters, they are unlikely to qualify.

• If a claim is rejected, penalties can be steep — the ATO generally seeks to impose a penalty equal to 50% of the tax shortfall, citing recklessness. The compliance stakes are high and rising.

• DISR's guidance requires that you compile contemporaneous documentation before commencing any experimentation, document a specific hypothesis, and describe how you intend to validate a technical or scientific challenge — documentation must be built in real time, not reconstructed after the fact.

Conclusion

The R&D Tax Incentive is one of the most valuable — and most misused — tools available to Australian AI companies. The 43.5% refundable offset for eligible small companies represents a material funding advantage, particularly for capital-intensive AI development involving model training, custom architecture design, and novel data pipeline engineering. But the program's self-assessment design, combined with increasingly active ATO and DISR compliance activity, means that poorly structured or inadequately documented claims carry serious financial risk.

The fundamental discipline required is straightforward: every AI project claimed under the R&DTI must be anchored to a genuine, documented technical uncertainty that could not be resolved by applying existing knowledge — and that uncertainty must be tested through a systematic, hypothesis-driven experimental process. Prompt engineering, routine model fine-tuning, and standard software integration do not meet this bar.

For businesses navigating the full landscape of Australian government AI support — from the R&DTI through to the AI Adopt Program, NAIC services, and research commercialisation grants — understanding how these programs interact, and where the compliance boundaries lie, is essential to building a sustainable innovation funding strategy. See our Australian Government AI Strategy: The Complete Guide to Grants, Programs and Business Support for the full picture.

References

• Australian Bureau of Statistics. "Research and Experimental Development, Businesses, Australia 2023–24." ABS, August 2025. https://www.abs.gov.au/media-centre/media-releases/ai-now-fastest-growing-area-business-rd

• Australian Taxation Office. "About the R&D Program." ATO, 2025. https://www.ato.gov.au/businesses-and-organisations/income-deductions-and-concessions/incentives-and-concessions/research-and-development-tax-incentive-and-concessions/research-and-development-tax-incentive/about-the-r-d-program

• Australian Taxation Office. "R&D Tax Incentive Registration Deadline Approaching." ATO, 2025. https://www.ato.gov.au/businesses-and-organisations/business-bulletins-newsroom/r-and-d-tax-incentive-registration-deadline-approaching

• Australian Taxation Office. "R&D Tax Incentive Transparency Reports." ATO, 2024–2025. https://www.ato.gov.au/businesses-and-organisations/income-deductions-and-concessions/incentives-and-concessions/research-and-development-tax-incentive-and-concessions/research-and-development-tax-incentive/r-d-tax-transparency-reports

• Austrade. "Australia Launches National AI Plan to Build a World-Class AI Industry." Australian Trade and Investment Commission, 2025. https://international.austrade.gov.au/en/news-and-analysis/news/australia-launches-national-ai-plan-to-build-a-world-class-ai-industry

• Australian Government. "Research and Development Tax Incentive." business.gov.au, 2025. https://business.gov.au/grants-and-programs/research-and-development-tax-incentive

• Department of Industry, Science and Resources. "R&D Tax Incentive: Guide to Interpretation." DISR / AusIndustry, updated June 2024. https://business.gov.au/-/media/grants-and-programs/rdti/rdti-guide-to-interpretation-2020-pdf

• Income Tax Assessment Act 1997 (Cth), Division 355, ss. 355-5, 355-25, 355-30. https://www5.austlii.edu.au/au/legis/cth/consol_act/itaa1997240/s355.25.html

• PwC Australia. "July 2025 R&D Tax Incentive Bulletin: Insights and Updates." PwC Australia, July 2025. https://www.pwc.com.au/pwc-private/r-and-d-gov-incentives/tax-incentives/r-and-d-tax-incentive-bulletin-july-2025.html

• PwC Australia. "A Quick Guide to R&D Tax Incentive Regulatory Reviews." PwC Australia, 2025. https://www.pwc.com.au/pwc-private/r-and-d-gov-incentives/tax-incentives/a-quick-guide-to-R-D-tax-incentive-regulatory-reviews.html

• RSM Australia. "R&D Tax Incentive Faces New Compliance Challenges." RSM Australia, August 2025. https://www.rsm.global/australia/insights/rd-tax-incentive-faces-new-compliance-challenges

• William Buck Australia. "Are Machine Learning and Artificial Intelligence Activities Eligible for the R&D Tax Incentive?" William Buck, May 2025. https://williambuck.com/news/em/technology/are-machine-learning-and-artificial-intelligence-activities-eligible-for-the-rd-tax-incentive/

• Swanson Reed. "Assessment of AI Related Activities Under the R&D Tax Incentive and Potentially Emerging Compliance Focuses." Swanson Reed, August 2024. https://www.swansonreed.com.au/assessment-of-ai-related-activities-under-the-rd-tax-incentive-and-potentially-emerging-compliance-focuses/