R&D Tax Incentive · Software & AI
R&D Tax Incentive for Software & AI Development
Software and AI development can qualify for the R&D Tax Incentive — but only the parts that involve genuine technical uncertainty resolved through systematic experimentation. Routine development — standard features, configuration, using libraries as intended — is not R&D. Because the ATO actively scrutinises software and AI claims, getting this distinction right (and documenting it) matters. Whether your specific activities qualify is self-assessed. See all industries.
On this page: what counts as software R&D · eligible vs ineligible activities · AI & machine learning · how an RSP helps · case study · who does what · FAQ
Not sure yet whether this is R&D? Start with the pathway
“We think AI could work on our own data — but we can't tell whether the model will actually perform, or whether we're just hoping.”
The uncertainty usually isn't the technology in general — it's whether the approach works on your data, at your accuracy threshold, under your constraints: the part no one has validated in your context before.
What counts as R&D in software & AI
The test is the same as for any R&D Tax Incentive claim: a core R&D activity is experimental work whose outcome cannot be known or determined in advance based on current knowledge, and which you resolve through a systematic progression from hypothesis to experiment, observation and conclusion. In software, that means the technical result was genuinely uncertain — not just unbuilt.
Supporting activities (e.g. building a test harness or dataset directly required for the experiment) may also be eligible where their dominant purpose is to support the core R&D. Commercial uncertainty — whether customers will buy it — is never technical uncertainty.
Eligible vs ineligible software activities
| May be eligible (if genuinely uncertain) | Generally not eligible |
|---|---|
| Developing a new algorithm whose behaviour is uncertain | Routine coding, bug fixing, configuration |
| Novel machine-learning model development & validation | UI changes; standard features (login, dashboards, CRUD) |
| Achieving performance/scalability not reachable with known methods | Using libraries, frameworks or APIs as intended |
| Integrating systems where no established method exists | A standard online quoting tool built with known methods |
These are general examples, not a ruling on your project. The same activity can be eligible in one context and not in another — it depends on whether real technical uncertainty exists.
AI & machine learning
AI and ML activities are judged on exactly the same test — the activity, not the buzzword. Developing and validating a novel model whose performance can't be predicted in advance may be eligible; calling an existing model or LLM API as intended is normal development. Using AI does not, by itself, make a project R&D, and the ATO actively reviews AI claims — so the technical-uncertainty test and good records are what protect a claim.
How a registered RSP helps software R&D
A Registered Research Service Provider conducts and structures your R&D — framing the hypothesis, designing the experiments, defining the metrics and keeping the evidence the ATO expects. For smaller software budgets there is a second point: spend with a non-associate registered RSP is what can still support a claim below the $20,000 threshold — other in-house costs do not automatically count toward it. See claiming R&D under $20,000 and what an RSP is.
Case study — AI & software systems
A company had an AI platform and automation tools in mind but an unclear technical path, and could not tell which parts were genuine experimental R&D versus normal feature development. Ignition helped break the system down into its goals, core technical challenges, algorithm options, data needs, benchmarks, performance metrics and validation methods.
R&D capability built: the company moved from a system concept to a complete software R&D framework — from experimental development through to performance validation — structured and documented to support a defensible R&DTI position (self-assessed, with the tax claim handled by their registered tax agent).
Discuss your software or AI R&D projectWho does what (the role boundary)
What we are
An external research office
A government-registered research organisation (RSP000047) that frames your problem, runs the investigation and keeps every step defensible.
Not an AI vendor
We don't sell you a build.
InsteadWe test whether your idea can hold up on real data before you commit to building it.
Not a tax agent
We don't promise claims or refunds.
InsteadWe conduct the research. Tax advisers prepare the claims.
Not a grant writer
We don't pitch funding outcomes.
InsteadWe help you see which support pathways may be relevant — and what preparation they need.
Ignition's role as a registered RSP is scientific, technical and research-structuring support — conducting and documenting the software R&D. Tax advice, the offset calculation, lodgement and your entity-specific claim position should be handled by your registered tax agent or accountant. We work alongside them. See RSP vs R&D tax consultant.
Frequently asked questions
Is software development eligible for the R&D Tax Incentive?
Sometimes — but only the parts that involve genuine technical uncertainty resolved through systematic experimentation. Building software with known methods and existing tools is normal development, not R&D. Where the technical outcome could not be known or determined in advance based on current knowledge, and you work it out through a systematic, experimental process, those activities may be eligible. Eligibility is self-assessed.
What software activities can qualify — and what generally does not?
May qualify: developing a new algorithm whose behaviour is uncertain, novel machine-learning model development, achieving performance or scalability that cannot be reached with known approaches, or integrating systems in ways no established method covers. Generally does not qualify: routine coding, bug fixing, configuration, UI changes, standard features (login, dashboards, CRUD, typical CRMs), and using libraries or platforms as intended. A standard online quoting tool built with known methods is the classic ineligible example.
What counts as "technical uncertainty" in software?
Technical uncertainty means a competent professional in the field could not have known or determined the outcome in advance from publicly available knowledge — so you have to experiment to find out. Commercial uncertainty (will it sell?), or simply not yet having built it, is not technical uncertainty. The uncertainty must be technical or scientific.
Are machine learning and AI activities eligible for the R&D Tax Incentive?
They can be, on the same test: AI/ML work may be eligible where you are resolving genuine technical uncertainty through systematic experimentation — for example developing and validating a novel model whose performance cannot be predicted in advance. The activity, not the buzzword, is what matters.
Does using AI or an LLM automatically make my project eligible?
No. Calling on AI, an LLM or any modern technology does not by itself make a project R&D. Using an existing model or API as intended is normal development. Eligibility depends on whether you are resolving technical uncertainty through systematic experimentation — and the ATO actively scrutinises software and AI claims, so the distinction matters.
Can my SaaS startup claim the R&D Tax Incentive if we spent under $20,000?
Possibly. You normally need at least $20,000 of total notional R&D deductions. Below that, the offset is worked out on a narrower base: broadly, expenditure to a registered Research Service Provider (RSP) that is not your associate, for services within a research field it is registered for, plus eligible contributions to a Cooperative Research Centre. Other in-house amounts — salaries, cloud costs, materials — do not automatically form part of that below-threshold base. So a smaller software R&D project may still be claimable when the experimental work is conducted through an RSP, subject to the activities being eligible (which you self-assess).
Why do software R&D claims get rejected, and what does the ATO look for?
Common reasons: claiming routine development as R&D, no identifiable technical uncertainty, no systematic experimentation, and weak contemporaneous records. The ATO looks for a clear hypothesis, the experiments run, the results, and evidence kept at the time. Structuring the R&D and its records properly up front is the best protection.
Can I claim if the software was developed overseas?
Generally R&D activities must be conducted in Australia to be eligible. Overseas R&D can only be claimed with an Overseas Finding from AusIndustry, granted in limited circumstances. Confirm your specific position with a registered tax agent.
Is Ignition Research a registered RSP for software R&D?
Yes — Ignition Research is a Registered Research Service Provider (RSP000047) and structures and conducts software and AI R&D for Australian companies. You can verify the current registration on the public RSP register at business.gov.au.
Sources
General information only — not tax, financial or legal advice. R&D Tax Incentive eligibility depends on your specific activities and circumstances and can change; whether software or AI work qualifies is self-assessed. The ATO scrutinises software and AI claims — always confirm your position with the Australian Government (business.gov.au and ato.gov.au) or a registered tax agent. Last reviewed: June 2026.