AI is reshaping how Australian businesses innovate – and that’s creating real opportunities with the R&D Tax Incentive (RDTI). But it’s also attracting closer scrutiny from the ATO and the Department of Industry, Science and Resources (DISR). If you’re using AI in your development work and thinking about claiming, here’s what you need to know.

How AI-related projects align with the RDTI

The RDTI rewards companies for conducting genuine experimental R&D – work where the outcome couldn’t be known in advance and that generates new knowledge through a systematic process. Many AI projects fit this description naturally.

Three criteria tend to apply well to AI development:

Advancement in technology – if you’re building custom models, developing new algorithms or pushing the boundaries of what existing tools can do, you may well be generating the kind of technical advancement the RDTI is designed to support.

Technical uncertainty – developing AI solutions often involves unpredictable outcomes. If you can’t know whether your model will perform as intended without running experiments, that uncertainty is a core part of what makes the work eligible.

Systematic experimentation – training, testing and refining AI models through iterative cycles maps closely to the hypothesis-experiment-evaluation process the RDTI requires. Document each cycle and you’re building your claim as you go.

The critical distinction is between genuine R&D and routine software development. Using an off-the-shelf AI tool to automate an existing process is not R&D. Building and iterating on a custom solution where the technical outcome is genuinely uncertain – that’s where eligibility starts.

Which industries are seeing the strongest AI-related claims?

Several Australian sectors are actively combining AI development with RDTI claims:

• Manufacturing – predictive maintenance, quality control and robotics applications
• Healthcare – diagnostic tools, patient monitoring systems and personalised treatment models
• Agriculture – machine learning for crop yield prediction, pest control and soil analysis
• Financial services – fraud detection, automated compliance systems and risk modelling

Across all of these, the same rules apply. The AI work needs to be genuinely experimental, technically uncertain and documented properly. The industry context doesn’t change the eligibility criteria – it just shapes how those criteria apply.

Why AI claims are getting more scrutiny

The ATO and DISR have increased their focus on AI-related RDTI claims, and it’s worth understanding why so you can anticipate what they’re looking for.

The three most common issues that trigger scrutiny are:

Blurred lines between R&D and routine development – AI projects can look like R&D from the outside but still fail the eligibility test if the technical outcomes were reasonably predictable using existing knowledge. Assessors are specifically looking at whether the uncertainty was genuine and technical, not just commercial.

Over-reliance on existing platforms – adapting or fine-tuning a third-party AI tool without generating new knowledge is not core R&D under the RDTI. If your project is primarily a configuration or integration exercise, the eligibility bar is much higher and the claimable portion of your costs may be significantly smaller than you expect.

Insufficient documentation – this is the most common failure point across all RDTI claims, but it’s especially acute for AI projects where the experimental process can be fast-moving and informal. The ATO expects to see contemporaneous records – documentation created at the time, not reconstructed after the fact.

What good documentation looks like for AI R&D

Strong documentation for an AI-related RDTI claim covers:

• The hypothesis you were testing – what were you trying to achieve and why wasn’t the outcome knowable in advance?
• The experimental process – what did you test, how did you test it and what happened?
• Failures and iterations – the RDTI specifically values negative results as evidence of genuine experimentation; don’t hide them
• The distinction between core R&D activities and any supporting or commercial work running alongside them
• Timesheets and cost records tied specifically to R&D activities, not general development work

Starting this documentation at the beginning of a project is far easier than reconstructing it at claim time. A weekly habit of recording what you tested, what happened and what you learned is enough to build a defensible record over a financial year.

What to do before you lodge a claim

Given the increased scrutiny on AI-related claims, getting an independent review of your eligibility before you lodge is worth the time. An R&D tax specialist can assess whether your activities genuinely meet the core R&D definition, identify which costs are claimable and flag any aspects of your claim that could attract an audit.

The RDTI can return up to 43.5% of your eligible R&D costs – for an AI-heavy startup, that can be a meaningful cash injection. But a claim that doesn’t hold up under review is worse than no claim at all.

Thinking about claiming the RDTI for your AI development work? Our R&D tax specialists work with Australian startups and scale-ups at every stage of the claims process – from eligibility assessment through to lodgement. Find out how our R&D Tax Incentive service works.