Software is one of the largest sources of R&D Tax Incentive claims in Australia, and also one of the most closely reviewed by AusIndustry and the ATO. The reason is simple: a lot of software development is genuinely experimental, and a lot of it is not. The program is designed to support the first kind and exclude the second, and the line between them is where claims succeed or fail.
This guide explains how the eligibility rules actually apply to software, with worked examples. It is written for founders, CTOs, and finance leads who want to understand what qualifies before they commit to a claim.
What are the four eligibility criteria?
A core R&D activity must be experimental, have an outcome that cannot be known in advance, follow a systematic progression from hypothesis to conclusion, and be conducted for the purpose of generating new knowledge. Under Division 355 of the ITAA 1997 an activity must meet all four tests; miss one and it is not core.
- Experimental in nature. The work is an experiment, not just construction.
- Outcome cannot be known in advance. This is tested against the "competent professional" standard with access to worldwide knowledge. If a skilled practitioner could have looked it up or reasonably predicted the result, it fails.
- Systematic progression of work. A clear path from hypothesis to experiment to observation to evaluation to conclusion.
- Purpose of generating new knowledge. The dominant purpose is to create new knowledge, not simply to build a product.
The ATO and AusIndustry have published a software development sector guide that is worth reading alongside this one.
Which software activities qualify?
Work qualifies when its core is a genuine technical unknown: a novel algorithm, an undocumented scaling regime, a capability the field has not demonstrated for your conditions. Work built with known methods, however difficult, does not. Here is how that lands on actual software work.
| Activity | Likely eligible? | Why |
|---|---|---|
| Designing a novel algorithm to hit a latency target no known approach achieves | Yes (core) | Genuine technical uncertainty, outcome unknown, requires experimentation |
| Building a standard CRUD web app with a popular framework | No | Known methods, outcome predictable for a competent professional |
| Developing a new approach to scale a database beyond documented limits for your workload | Yes (core) | Uncertainty about whether and how it can be done |
| Integrating a third-party API per its documentation | No | Routine, outcome knowable in advance |
| Experimenting with a new ML model architecture where performance is genuinely unknown | Often (core) | Depends on whether the uncertainty is real, not just tuning |
| Fixing bugs, refactoring, or routine maintenance | No | Not experimental, no new knowledge |
The pattern: eligibility tracks technical uncertainty, not effort, cost, or commercial importance. A difficult, expensive, business-critical feature built with known methods is still not R&D. A small, scrappy experiment into something genuinely unknown can be.
The agile trap
Agile, waterfall, or any other methodology does not, by itself, make work eligible or ineligible. AusIndustry has been explicit about this. The risk with agile specifically is that the ceremony of sprints, tickets, and standups can look like systematic experimentation when it is really just iterative delivery.
Tickets are not a hypothesis
A sprint board shows what you built and when. It does not show the technical uncertainty you faced, the hypothesis you formed, or how you evaluated the result. Those are the things a reviewer wants to see. Map your experiments to the work, do not assume the work speaks for itself.
The fix is not to abandon agile. It is to capture, alongside your normal process, the experimental story: what was uncertain, what you hypothesised, what you tried, and what you learned. That is also exactly what contemporaneous evidence requires, which we cover in our evidence and documentation guide.
AI and machine learning: more reward, more scrutiny
AI development is a rich source of genuine R&D, and it is attracting greater attention from the ATO and the Department of Industry, Science and Resources. Training a model on a documented dataset with established techniques to a predictable result is not automatically R&D. Pushing into genuinely uncertain territory, where a competent professional could not predict whether your approach will work, can be. The distinction is the same as everywhere else, but the volume of AI claims means the framing needs to be especially clean.
What costs can you claim?
For an eligible activity you can typically claim the salaries of staff directly engaged in the R&D (apportioned to their R&D time), contractor costs, materials consumed, and a share of directly attributable overheads. The expenditure you can include typically covers:
- Salaries of staff directly engaged in the R&D, including employer superannuation, apportioned to their R&D time.
- Contractor costs for R&D services from unrelated parties.
- Materials consumed in the R&D.
- A share of overheads directly attributable to the R&D.
Commonly excluded: management time not directly in R&D, sales and marketing, market research, and routine development using known methods. For software, the single biggest determinant of claim size is usually how staff R&D time is measured and substantiated, which is why time records matter so much.
Who can claim the R&D Tax Incentive?
Australian companies (not sole traders or partnerships) that conducted eligible R&D activities and register them with AusIndustry on time. The basics:
- You must be a company, and generally you should own or have appropriate rights to the resulting IP.
- The work must meet the four criteria above, self-assessed.
- You register your activities with AusIndustry within 10 months of the end of your income year, then claim the offset in your company tax return with the ATO.
- Today the offset is 43.5% refundable for companies with aggregated turnover under $20m (company tax rate plus 18.5 percentage points for a base-rate entity). At $20m turnover or more the offset is non-refundable, set at the company tax rate plus a two-tier R&D intensity premium: 8.5 points on notional R&D up to 2% of total expenses and 16.5 points on the portion above. There has been no flat non-refundable rate since 1 July 2021. Note these settings are changing: see our explainer on the 2026-27 Budget changes from 1 July 2028.
Self-assessment means the responsibility is yours
The R&DTI is a self-assessment program. AusIndustry and the ATO can review, audit, or amend any claim. That is not a reason to be timid, it is a reason to be accurate and well-documented. Rand assists with preparing and documenting your claim, but you and your directors remain responsible for it, and you should seek advice specific to your situation.
Where to go next
If you have read this far and think you have genuine R&D, the next question is evidence. Eligibility gets you in the door, contemporaneous records keep you there. Read how to document a defensible claim next, or step back to the end-to-end guide to the whole program for offset rates, registration, and lodgement.