R&D Tax Incentive for AI and Software Development: Is Your Algorithm Eligible?

Australian software companies, SaaS platforms, fintech providers, healthtech businesses, agtech innovators and AI developers are building products that solve genuinely complex problems. Behind many of these digital products are algorithms, machine learning models, data science tools, automated decision-making systems and advanced software engineering methods that go well beyond routine coding.

For Australian businesses, the R&D Tax Incentive may provide meaningful financial support where software development involves genuine experimentation, technical uncertainty and the creation of new knowledge. But not every software project will qualify. The central question is whether the work goes beyond ordinary development and involves eligible R&D activities under Australian law.

This article examines software development R&D tax claims, algorithm R&D tax incentive eligibility, machine learning R&D tax incentive considerations and the practical records Australian businesses should maintain when assessing whether their AI or software project may qualify.

 

Why the R&D Tax Incentive Matters for Australian Software Companies

The Australian R&D Tax Incentive is designed to encourage businesses to invest in genuine innovation. For software companies, this can be especially relevant because many digital products require sustained technical development before they become commercially viable.

Eligible companies with an aggregated turnover under $20 million can access a 43.5% refundable tax offset. Larger companies access a non-refundable offset of between 33.5% and 46.5%, depending on their corporate tax rate and the proportion of R&D expenditure to total business expenditure. Where software or AI development involves genuine technical uncertainty and experimental activities, these offsets can meaningfully support investment in talent, infrastructure and product development. For further detail on eligibility thresholds and offset rates, see the ATO’s R&D Tax Incentive guidance.

The incentive is especially relevant for start-ups and scale-ups investing heavily in development before reaching profitability. It is an entitlement programme, not a competitive grant, which means you do not need to outcompete other applicants. You need only meet the criteria.

 

What Software Development R&D Actually Means in Australia

Software development R&D is not simply about writing code. It is about whether the development process involves a technical problem that cannot be resolved using existing knowledge, standard tools or routine engineering alone.

Building a basic application using known frameworks is ordinary development. Designing a new algorithm that must process large volumes of data in a novel way, or testing whether a machine learning model can achieve a technically uncertain performance outcome, may involve eligible R&D activities in software engineering. The distinction matters. Australian software businesses need to identify the experimental work within the broader software development lifecycle. The R&D component may only be part of a total project, but it can still be significant if it involves hypothesis testing, technical risk and systematic experimentation.

One common mistake is treating the entire development project as R&D simply because the end product is new. The ATO and AusIndustry have both flagged this as a concern. Only the genuinely experimental components of a project should be registered.

 

Is Your Algorithm Eligible?

An algorithm may be relevant to an R&D Tax Incentive claim when it forms part of an experimental process aimed at resolving technical uncertainty. The focus is not simply whether the algorithm is commercially useful, but whether the development team had to test, evaluate and refine a technical approach where the outcome could not be known in advance.

Algorithm R&D tax incentive eligibility can arise in projects involving machine learning models, predictive analytics, natural language processing, computer vision, neural networks, optimisation engines, data classification systems or automated decision-making software.

Consider a company developing a predictive analytics algorithm. They may need to test different model architectures, training methods, data structures and performance thresholds before knowing whether the intended technical outcome is achievable. If that work is systematic, properly documented and directed at generating new knowledge, it may support an eligible R&D Tax Incentive assessment. The key is that the team genuinely did not know whether the approach would work until they tested it.

 

Technical Uncertainty in Software Development

Technical uncertainty sits at the heart of most software R&D claims. It exists where a competent professional cannot determine whether a technical outcome is achievable, or how it can be achieved, based on existing knowledge and standard practice.

In software development, technical uncertainty may relate to speed, accuracy, scalability, latency, interoperability, security, automation, data quality, model performance or system architecture. For AI and machine learning projects, uncertainty may also involve whether a model can be trained to achieve a required level of reliability across varied real-world data.

Document this uncertainty from the start of the project. Businesses should be able to explain what was unknown, why it could not be resolved through ordinary development, what experiments were conducted and what knowledge was gained. Records created after the fact are harder to defend and carry less weight with both AusIndustry and the ATO.

 

Core R&D Activities for Software and AI Projects

Core R&D activities are the experimental activities carried out to generate new knowledge and resolve technical uncertainty. In software and AI development, these may include testing a new algorithmic method, developing an experimental software architecture, trialling a novel machine learning model or evaluating whether a system can perform under difficult technical conditions.

Core R&D activity eligibility depends on the specific facts of the project. Each activity should involve a hypothesis, a structured experiment, observation, evaluation and logical conclusions. This is particularly important for projects involving experimental algorithms where the development team is testing whether a new technical approach can work, rather than applying a known solution.

Clearly separating core R&D activities from ordinary commercial development is one of the most important steps in preparing a defensible claim. The software development sector guide on business.gov.au provides useful self-assessment guidance for software businesses working through this distinction.

 

Supporting R&D Activities in Software Engineering

Supporting R&D activities are those that directly support core R&D activities. In software projects, these may include preparing test data, building experimental environments, running simulations, analysing model results, developing test harnesses or documenting technical findings.

Supporting activities need a clear connection to the core R&D work to be eligible. For example, data cleaning may not be eligible on its own if it is routine. But it may support eligible R&D where it is necessary to prepare a dataset for testing a machine learning model under experimental conditions. The connection matters and should be explained in the claim.

This distinction helps businesses identify which parts of a project may be eligible and which are ordinary development or commercial implementation work.

 

Machine Learning and AI R&D in Practice

Machine learning and artificial intelligence projects often involve genuine experimentation, particularly where businesses are trying to improve accuracy, reduce error rates, automate complex decisions or process unstructured data in a new way.

Machine learning R&D tax incentive projects may involve testing different training approaches, model structures, feature engineering techniques, data pipelines or validation methods. AI algorithm R&D may also be relevant where the business is developing new or improved methods for classification, recommendation, prediction, natural language processing or computer vision.

Using an existing AI model or applying standard software tools will not automatically qualify. The business must show that the work involved technical uncertainty, a systematic experimental approach and the creation of new knowledge. The method of experimentation matters as much as the technology being developed.

 

Examples of Software Projects That May Include Eligible R&D

Many Australian software projects contain eligible R&D components when they involve genuine technical experimentation. A SaaS business may be developing a new platform architecture to handle unusually complex workloads. A fintech company may be testing a novel risk assessment algorithm. A healthtech business may be building software to interpret clinical data more accurately. An agtech provider may be developing predictive tools using environmental and production data.

Other potential examples include cybersecurity software involving novel detection methods, cloud computing projects testing new distributed architectures, deep learning R&D aimed at improved classification performance, computer vision applications for industrial inspection, and data science tools that require new approaches to processing unstructured data.

Eligibility depends on the technical work, not the industry label. A project in fintech, healthtech, agtech or SaaS may be eligible if it involves experimental activities directed at resolving genuine technical uncertainty.

 

Software Activities That Are Less Likely to Qualify

Not all software development is eligible R&D. Routine coding, standard application development, ordinary website builds, user interface updates, minor feature improvements, bug fixing, maintenance, standard integrations and basic configuration work are generally not eligible where they do not involve technical uncertainty.

Assessing eligible versus ordinary software development activities carefully is essential before preparing a claim. A commercial software R&D tax offset claim should not be based solely on the fact that a product is new to the business or useful to customers. The work must involve experimental activities that seek to resolve technical uncertainty about whether a specific outcome is achievable.

This does not mean commercial software can never form part of an R&D claim. It means the business must identify the experimental components within the broader commercial project and support them with clear, contemporaneous evidence.

 

Documentation and Record Keeping for Software R&D

Good records are essential for software R&D tax claims. Australian businesses should keep contemporaneous records explaining the technical uncertainty, the hypothesis being tested, the experiments conducted, the results observed and the conclusions reached. As Bentleys’ R&D Tax Incentive eligibility guidance notes, evidence should clearly show that work proceeded from hypothesis to experimentation, observation and logical conclusions.

Useful records include project plans, technical specifications, source code repositories, sprint notes, issue tracking records, test results, model training logs, architecture diagrams, prototype documentation, developer notes and experiment summaries.

Build documentation practices into the development process rather than reconstructing records afterwards. Project teams in software and AI move quickly, and records created at the time carry significantly more weight than notes assembled after the financial year ends.

 

Source Code, Testing and Prototypes

Source code can be valuable evidence in a software R&D claim, but it should not be the only record. Source code R&D tax incentive eligibility is stronger when supported by records that explain why certain technical decisions were made, what experiments were performed and what was learned from each iteration.

Software testing for R&D purposes should focus on experimentation rather than routine quality assurance. Standard bug-finding and regression testing may be ordinary development. Testing designed to evaluate a technical hypothesis, measure experimental performance or compare uncertain technical approaches may be more relevant to the claim.

Prototypes can also play a role where they are used to test whether a technical outcome can be achieved before full commercial implementation. Document the purpose and findings of each prototype clearly.

 

Eligible Software Expenditure

Eligible expenditure connected to software R&D claims may include staff salaries, contractor costs, cloud computing costs and testing expenses, where these are directly connected to eligible R&D activities. The specific treatment depends on the facts of each project.

Because software projects often mix eligible and non-eligible activities, businesses should use a reasonable and defensible method to separate R&D expenditure from ordinary commercial development costs. The separation method should itself be documented. Qualifying R&D expenditure should be traceable to specific registered activities, not allocated in a broad or approximate way.

 

AusIndustry, the ATO and Australian Software Guidance

The R&D Tax Incentive is jointly administered by the Department of Industry, Science and Resources (through AusIndustry) and the Australian Taxation Office. Software companies should review the official AusIndustry software development guidance and ATO requirements when assessing their projects.

AusIndustry’s guidance is particularly relevant because software projects often mix experimental development, routine coding and commercial implementation. The updated software guide, last revised in 2024, clarifies when software activities excluded as core R&D activities may still qualify as supporting R&D activities. Australian companies should use Australian guidance and work with advisers familiar with the Australian programme. The rules differ meaningfully from overseas R&D tax credit systems and should not be confused with them.

 

Common Mistakes in Software R&D Claims

Assuming that all software development is R&D because the product is new or technically sophisticated is a consistent error. Another common problem is describing the commercial goal without clearly explaining the technical uncertainty. Businesses may also fail to separate eligible R&D activities from ordinary development, or may keep records that are too thin to support a detailed review.

Technical risk considerations can arise where a claim is too broad, poorly documented or focused on business uncertainty rather than technical uncertainty. Market risk, customer adoption risk and commercial viability are not technical uncertainties for R&D purposes. The uncertainty must relate to whether a specific technical outcome can be achieved and how.

Software businesses can strengthen their position by clearly identifying the experimental work, documenting the systematic progression followed and linking expenditure to specific eligible activities. Early preparation matters far more than a last-minute push before lodgement.

 

Why Specialist Advice Matters

R&D tax incentive claims in the software and AI sector are often complex. The technical work may be fast-moving, iterative and difficult to translate into the language required for registration and assessment. A specialist adviser can bridge the gap between engineering processes and R&D claim requirements.

Working with Bentleys’ R&D Tax Advisory team gives software and AI businesses access to advisers experienced across fintech, SaaS, healthtech, agtech and technology-based industries. Support can include identifying eligible activities, reviewing documentation, assessing supporting activities, apportioning expenditure and preparing a registration that reflects Australian R&D Tax Incentive requirements.

For founders, CTOs, CFOs and business owners, advice is most valuable before the claim is prepared. Early engagement helps development teams improve record keeping, define technical uncertainties more precisely and avoid treating ordinary commercial development as eligible R&D.

 

Getting Started

Australia’s AI and software sector contains businesses developing serious technology, from machine learning platforms and data science tools to SaaS products, cybersecurity systems, predictive analytics engines and novel software architectures. For companies genuinely experimenting to resolve technical uncertainty, the R&D Tax Incentive can provide substantial support.

The key is careful, honest assessment. A well-prepared claim should be grounded in experimental activities, technical uncertainty, systematic testing, new knowledge and strong contemporaneous records. It should not rest only on the fact that software is new, commercial or technically impressive from a market perspective.

Review your current and recent projects. Identify where technical uncertainty existed. Gather your records. Then speak with specialist Australian advisers through Bentleys’ chartered accountants and business advisory team before making a claim. The R&D Tax Incentive rewards genuine innovation,  but only when it is documented and presented correctly.

 

Disclaimer: This information is general in nature and should not be relied on as advice. It does not take into account the objectives, financial situation or needs of any particular person. You need to consider your financial situation and needs and seek professional advice before making any decisions based on this information.

 

FAQs

Can an AI algorithm qualify for the Australian R&D Tax Incentive?

Yes, an artificial intelligence algorithm can qualify, provided it is developed through a systematic progression of work aimed at resolving a specific technical uncertainty. The project must aim to generate new knowledge rather than simply applying existing machine learning models or libraries to a standard business problem.

Who administers the R&D Tax Incentive for software companies in Australia?

The program is jointly administered by two federal bodies. AusIndustry (under the Department of Industry, Science and Resources) assesses the eligibility of the technical software activities, whilst the Australian Taxation Office (ATO) manages the financial claims, eligible expenditure, and tax offset payouts.

What is the difference between core and supporting R&D activities in software development?

Core R&D activities are experimental processes that require a hypothesis, experimentation, and evaluation to resolve a technical uncertainty. Supporting R&D activities are distinct tasks that do not involve experiments themselves but directly relate to and support the core experimental software work.

What are the minimum expenditure requirements to claim the R&D tax offset in Australia?

An eligible corporate entity must incur at least $20,000 in qualifying R&D expenditure during the financial year to lodge a claim. This minimum spending threshold is waived if the experimental software development is contracted to a registered Research Service Provider (RSP).

Can a sole trader or partnership claim the R&D Tax Incentive for a new application?

No, individuals, sole traders, and corporate limited partnerships are strictly ineligible for the program. To claim the tax offset, the software development must be undertaken by an eligible R&D entity, which generally means a company incorporated under Australian law.

What tax offset rates apply to Australian tech startups under the R&D incentive?

For companies with an aggregated annual turnover of less than $20 million, a refundable tax offset is available, which provides a cash refund if the company is in a tax loss position. For larger entities with a turnover of $20 million or more, a non-refundable tax offset is applied to reduce corporate tax liability.

Are standard machine learning model integrations eligible for the tax offset?

Routine integration of pre-existing machine learning models using standard APIs or public libraries is generally deemed ineligible. To qualify, your team must be solving a deep technical challenge, such as optimizing a neural network architecture or training an algorithm to overcome a documented processing limitation, where the outcome cannot be determined in advance by a competent professional.

What software activities are explicitly excluded from being core R&D activities?

The Industry Research and Development Act explicitly excludes several software activities from being core R&D. These include cosmetic UI/UX design, routine bug fixes, data migration testing, market research, and software developed for the sole or dominant purpose of internal business administration.

How does the ATO define a “competent professional” in software engineering?

A competent professional is someone with relevant qualifications, skills, and industry experience who stays up to date with global developments in software engineering. The ATO uses this benchmark to determine if a technical problem truly required an experimental process or if it could have been solved using widely available knowledge.

What is an ATO Taxpayer Alert, and why does it matter for software claims?

The ATO issues Taxpayer Alerts, such as TA 2017/5, to warn businesses against higher-risk tax arrangements. For software developers, these alerts outline common compliance errors, such as claiming an entire commercial software project as R&D rather than isolating the specific experimental components.

Can internal administration software ever be claimed under the R&D program?

Software developed for the dominant purpose of internal business administration is excluded from being a core R&D activity. However, if an internal tool requires a highly innovative algorithm to function, that specific component may be claimed as a supporting R&D activity, provided it meets strict legislative criteria.

What kind of records must Australian software companies keep to support a claim?

Companies must maintain detailed, contemporaneous records generated at the time the software development occurred. This includes source code repositories (such as Git logs), project specifications, architecture diagrams, testing logs, hypothesis documentation, and precise timesheets tracking developer hours spent on experimental tasks.

What is the deadline for registering software R&D activities with AusIndustry?

You must register your R&D activities with AusIndustry within 10 months of the end of your company’s income year. For most Australian companies with a standard financial year ending 30 June, the strict registration deadline is 30 April of the following calendar year.

Is commercial risk enough to justify an R&D tax claim for a new app?

No, commercial risk, market uncertainty, or the risk that a product might not sell do not qualify a project for the incentive. The project must be driven by technical uncertainty, meaning that the software engineers do not know if the technical objective is achievable based on current global knowledge.

Can overseas software development costs be claimed under the Australian incentive?

Generally, only R&D activities conducted within Australia are eligible. To claim expenses for software development activities performed overseas, you must apply for and receive an Advance Finding from AusIndustry before incurring the expenditure, proving the work cannot be conducted locally.

What happens if a software company registers its R&D but gets audited later?

AusIndustry registration does not automatically mean a claim is fully approved or immune from future review. Both AusIndustry and the ATO conduct retrospective reviews and audits, sometimes years after a refund has been issued, to verify that the activities were genuinely experimental and that the expenditure was recorded correctly.

Does upgrading an existing software platform qualify for the tax offset?

Simply adding standard features, modifying existing code, or customising a commercial off-the-shelf software package does not qualify. The upgrade must involve a systematic process to overcome a significant technological limitation, such as re-engineering an algorithm to achieve unprecedented data processing speeds that existing frameworks cannot support.

Can deep learning or neural network optimization be considered core R&D?

Yes, if your data science team is creating a novel neural network architecture or adapting deep learning models to handle data in ways that push past current technological boundaries, these activities satisfy the criteria for core R&D due to the inherent algorithmic uncertainties.

Are software testing and prototyping covered by the R&D Tax Incentive?

Software prototyping and testing can be claimed, but their eligibility depends on their purpose. Routine system testing or user acceptance testing (UAT) is excluded, whereas building a prototype to test a specific algorithmic hypothesis and resolve a technical unknown is highly eligible.

Can web development or e-commerce platform creation be claimed?

Building standard websites, e-commerce platforms, or mobile apps using established content management systems and frameworks is considered ordinary business activity and is ineligible. Only the specific components that require experimental software engineering to solve a core technical bottleneck can be registered.

What is a systematic progression of work in the context of software R&D?

A systematic progression of work mirrors the scientific method within software development. It requires the formulation of a clear hypothesis (a predicted technical outcome), an experimental phase (writing and compiling experimental code), an evaluation phase (testing and measuring the results against the hypothesis), and a conclusion that yields new knowledge.

Can SaaS startups claim developer salaries under the R&D Tax Incentive?

Yes, the salaries of internal software developers can be included as eligible expenditure, but only for the proportion of time they spend directly working on registered core or supporting R&D activities. General maintenance, client onboarding, and routine DevOps work must be excluded from the calculation.

How does a company claim the R&D tax offset once registered with AusIndustry?

After AusIndustry approves your registration, they will issue a unique R&D registration number. Your tax agent or accountant must then complete and lodge an R&D Tax Incentive Schedule alongside your annual Australian company tax return to the ATO.

Can data science and predictive analytics projects qualify for government support?

Data science initiatives frequently qualify if they involve creating custom predictive analytics models that require new algorithmic solutions. If the project relies entirely on running clean data through standard, out-of-the-box analytical tools, it will not meet the requirement for technical experimentation.

What are the consequences of making an incorrect software R&D claim to the ATO?

If an audit reveals that ineligible ordinary business activities were claimed, the ATO will disallow the deductions. This results in the company having to repay the refunded amounts, plus potential shortfall penalties and compounding interest charges.