Quick answer: Neither party claims automatically. Section 355-210 of the ITAA 1997 lets an R&D entity notionally deduct expenditure only where the activity was conducted for it, and s 355-210(2) takes an activity outside that section where it is conducted "to a significant extent, for one or more other entities". The ATO reads "for" by asking who receives the major benefit — effective ownership of the results, control of the work, and financial risk. The same code can sit with either party on different facts. You self-assess.
18 August 2026 — this article describes the current rules. The 2026-27 Federal Budget announced R&DTI changes; the ATO states the measure is not yet law, and the Department of Industry, Science and Resources states the changes will apply to income years starting on or after 1 July 2028.
An Adelaide integrator spends four months on a model nobody in the building had built before, invoices the client, and then finds the live question is not whether the work was experimental. It is whose experiment it was. That question sits under almost every contracted AI or automation build, and both parties are routinely told — separately and confidently — that they can claim it.
Scope of this article: This piece is about who may claim a contracted AI or automation activity, not about which AI activities are eligible in the first place. Where the experimental line falls inside an AI project is covered in separate Ignition Research Insights on AI quoting engines and on LLM and RAG reliability, and more broadly in R&D for software and AI. Everything below assumes an activity already assessed against the eligibility tests.
The Test the Whole Question Turns On
Take eligibility first: "who claims" only arises once something is claimable. Core R&D activities are experimental 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 that is based on principles of established science and proceeds from hypothesis to experiment, observation and evaluation and leads to logical conclusions — and 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 (business.gov.au).
The established-science limb is where contracted software work: a disciplined sprint cadence with metrics on a dashboard is a progression of work, but not by itself one based on principles of established science proceeding to logical conclusions. AusIndustry's AI guidance makes the adjacent point plainly: "Using an AI model or technique that is new to you does not, by itself, mean the activity is eligible for the program", and "using AI in software development does not make an activity eligible" (business.gov.au).
"Conducted for" Is a Separate Question from "Conducted by"
An R&D entity can notionally deduct expenditure on registered R&D activities only where the conditions in s 355-210 are met. The ordinary domestic paragraph, s 355-210(1)(a), applies where the R&D activity is conducted for the R&D entity solely within Australia. Then the carve-out, verbatim:
"However, an R&D activity is not an activity to which this section applies if the activity is conducted, to a significant extent, for one or more other entities not covered by any paragraph of subsection (1)." — ITAA 1997, s 355-210(2)
Two things follow. Whose staff ran the trials decides nothing — R&D is routinely carried out under contract by people with expertise the claimant does not have. And the language is "to a significant extent", not "solely" or "mainly": an activity can be conducted for you in part and still fall outside the section because it was also conducted, to a significant extent, for someone else. How much is significant is a question of fact and degree on the whole of the circumstances.
The ATO's three indicia
"Working out for whom the R&D activities are conducted involves determining who receives the major benefit from carrying out the activities, for example, who effectively owns the results of the activities, controls the conduct of the R&D activities and bears the financial risk." — ATO, Who R&D activities are conducted for
Effective ownership, control and financial risk are indicia that together inform one question. They are not three tests each of which must be passed, and no one of them settles it.
It can genuinely go either way
The ATO's material includes an example in which the supplier is the entity the activities were conducted for: a company contracting to supply a product meeting a customer's specifications, under no obligation to supply working papers or background research, which alone controls and uses the results and effectively owns them — even where the customer is its only realistic buyer (ATO). Being the vendor does not put you outside the incentive any more than being the payer puts the client inside it.
Why the contract and the commercial substance both matter: The agreement can be important evidence: it is where ownership of foreground results, decision rights and payment risk are often documented. But the ATO's guidance directs attention to what practical, as well as formal, rights a party has to the results — so a signed IP assignment is evidence of where the benefit lands, not a substitute for looking at where it landed. A contract silent on the research record does not make the work the developer's by default either.
What the engagement looks like
Tends to point toward the client
Tends to point toward the developer
Who fixed the technical target and the acceptance measure
The client's team set and signed off the measure
The developer set it for its own product roadmap
Who decides to abandon a line of investigation
Client approval required to change direction
Developer decides and absorbs the consequence
Foreground results, method and evaluation harness
Assigned to the client, including working papers
Retained by the developer and reusable elsewhere
Payment
Milestones payable regardless of the technical result
Developer-funded, recovered only if the method works
Who carries the consequence of failure
Client's operational exposure is what the work addresses
Developer wears the cost of the failed trials
None of these rows is decisive alone, and real engagements produce mixed rows. Mixed rows are exactly the circumstance s 355-210(2) is aimed at.
Each entity's entitlement must be assessed at the activity level: Section 355-210 can prevent duplication of claims in some circumstances, but it is not a blanket rule that only one entity can ever have expenditure relating to the same R&D activity. Each R&D entity must establish that the relevant activity was conducted for it and was not conducted, to a significant extent, for disqualifying other entities. In a contracted engagement, the analysis therefore needs to be made at the activity level rather than simply by reference to the project, contract or invoice.
The Delivery Half of a Delivery Contract
Much of a routine contracted build may be delivery work where the relevant technical outcomes can be determined in advance, and what is unknown is effort rather than outcome. AusIndustry lists activities that are usually not core R&D activities, and several describe ordinary delivery directly: running regression, acceptance or functionality tests using established testing methods to confirm a system works as intended, where expected outcomes are already known; and integrating known model outputs such as scores, labels and recommendations into applications or dashboards using pre-defined logic, rules and interfaces.
Delivery work does not automatically attach as supporting activity either. Supporting R&D activities are activities directly related to core R&D activities — but where an activity is of a kind referred to in s 355-25(2), or produces goods or services, or is directly related to producing goods or services, it is a supporting R&D activity only if it is undertaken for the dominant purpose of supporting core R&D activities (business.gov.au). Where a particular delivery activity produces, or is directly related to producing, goods or services, the additional dominant-purpose test applies if that activity is being assessed as supporting R&D. Either way the dominant-purpose hurdle applies.
A Worked Example: An Adelaide Software Firm and a Demand-Forecasting Build
Hypothetical and illustrative only. Not a client, not a ruling, and not a statement that these facts would be eligible.
A 22-person Adelaide software firm is engaged by a national industrial-parts wholesaler to deliver a demand-forecasting capability into the client's ERP: data pipeline, forecast service, planner interface, and integration to replenishment. Fixed price $520,000 over 26 weeks. The catalogue is 41,000 SKUs across six distribution centres.
Baseline, fixed in week 2: The incumbent method is a 13-week moving average with planner override. On a temporal holdout of the most recent 26 weeks: weighted MAPE of 22% across the catalogue, but 61% on the intermittent tail — the 9,400 SKUs with four or fewer non-zero demand weeks per quarter, carrying 3% of revenue and 38% of stockout events.
Target, fixed in week 3 before the holdout was sealed: Mean absolute scaled error at or below 0.85 against a seasonal-naive benchmark on the intermittent tail, without degrading the fast-moving segment below its baseline.
Held constant & varied: Held constant: The SKU set, the holdout weeks, the error metric, the SKU-week aggregation level, and the promotion calendar as an exogenous input. Varied: The estimator family and the treatment of zero-inflation.
Trial 1: Croston's method and its Syntetos-Boylan variant — established intermittent-demand estimators. MASE 1.02: no better than seasonal naive on the tail. This ruled out the proposition that classical intermittent-demand estimators recover enough signal from this history.
Trial 2 — failed, and the informative one: A single gradient-boosted model trained across all 41,000 series, on the hypothesis that cross-series pooling would transfer information from fast movers to the tail. MASE 0.71 on the holdout — apparently a pass. A leakage audit found two features, weeks since last replenishment order and current on-hand, were read from a nightly snapshot that carried post-period values for the holdout weeks. Rebuilt point-in-time, the same configuration returned MASE 0.98. The trial failed, and it ruled out cross-series pooling on header features as the source of the gain: the improvement was information that would not exist at forecast time.
Trial 3: Hierarchical reconciliation — forecast at product-family × distribution-centre level, where demand is dense, then disaggregate to SKU on a rolling 52-week proportion. MASE 0.88 on the tail overall; 0.79 on the 5,100 tail SKUs whose family had at least 20 non-zero weeks.
Result reached: The target was not met across the intermittent tail as defined; it was met on a bounded subpopulation. The knowledge generated was that the binding constraint is demand density at the family level, not the choice of estimator — a different finding from the one the engagement had assumed.
Activity boundary: For this example, the candidate experimental activity is documented from the point at which the hypothesis and experimental approach are established through to the trials, evaluation and recorded conclusion. The pipeline, planner interface, ERP integration, deployment, UAT and training sit outside it.
The same work, two fact patterns:
As it actually ran (points to client): The client's supply-chain lead set the target measure, approved abandoning trial 2 and accepted the bounded result. The contract assigns all foreground results and working papers to the client, milestones are payable whether or not MASE 0.85 was reached, and the exposure addressed — stockouts in the client's network — is the client's. On those facts the indicia point to the client, and for the software firm the activity is generally unlikely to be one to which s 355-210 applies, because it was conducted to a significant extent for another entity, subject to the activity's own facts and the statutory tests.
A different engagement, same code (points to developer): The firm carries the forecasting investigation at its own cost outside the fixed-price scope, is not obliged to hand over working papers, decides itself which trials to run and when to stop, and keeps the reconciliation method and evaluation harness for later distributors. The client receives a deliverable meeting a specification. On those facts the indicia can point to the software firm — the pattern in the ATO's supplier example above. Neither version is a conclusion; both are self-assessed, and expenditure treatment is a separate question for the company and its registered tax agent.
Where an RSP Fits
business.gov.au describes Research Service Providers as scientific or technical service providers, registered in specific fields, that a company can engage to conduct R&D activities on its behalf (business.gov.au). Qualifying expenditure incurred to a non-associate RSP may still be taken into account in determining R&D tax offset entitlement where total notional deductions are below the usual $20,000 threshold, provided the services relate to a research field for which the RSP is registered. Using an RSP does not guarantee eligibility — you still self-assess. An RSP supplies research capability, not tax advice. Offset rates, the refundable and non-refundable tiers and the turnover conditions are covered separately in refundable vs non-refundable offset.
Frequently Asked Questions
Q: Can a software development agency claim the R&D Tax Incentive for client work?
A: Sometimes, but not because it wrote the code. Section 355-210 asks who the activity was conducted for, and s 355-210(2) takes an activity outside the section where it is conducted "to a significant extent, for one or more other entities". Where the client sets the technical target, owns the results and carries the risk, the activity is generally unlikely to be one the developer can claim on those facts, subject to the activity's own facts and the statutory tests.
Q: Who claims the R&DTI when the client pays for the build?
A: Paying for the work is one indicium, not the answer. The ATO asks who receives the major benefit, informed by who effectively owns the results, who controls the conduct of the activities and who bears the financial risk. A supplier who is not obliged to hand over its working papers, controls the investigation and reuses the results can be the entity the activities were conducted for, even where the customer funded the engagement. Each activity is self-assessed on its own facts.
Q: What does "conducted to a significant extent for one or more other entities" mean?
A: It is the carve-out in s 355-210(2) of the ITAA 1997. The activity does not have to be conducted entirely for someone else — the statutory language is "to a significant extent", so an activity conducted partly for you and, to a significant extent, for another entity can fall outside the section. How much is significant is a question of fact and degree assessed on the whole of the circumstances.
Q: Can both the developer and the client claim the same R&D activity?
A: Potentially, depending on the structure and facts. Section 355-210 can prevent duplication in some circumstances, but it does not impose a blanket one-claimant-per-activity rule. Each entity's entitlement must be assessed separately against the relevant registration, expenditure and “conducted for” requirements.
Sources & Further Reading
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25, 355-30 and 355-210
ATO — Tax Reform: better targeting the R&D Tax Incentive — announced 12 May 2026; the ATO states the measure is not yet law
industry.gov.au — Research and Development Tax Incentive — states the changes apply to income years starting on or after 1 July 2028
Related: R&D for software and AI · what does not qualify · what an RSP is · refundable vs non-refundable offset · R&D Tax Incentive in Adelaide · Insights
Talk to Ignition Research before the statement of work is signed. As a Registered Research Service Provider at Lot Fourteen in Adelaide, we assist with experimental design, technical R&D work and contemporaneous supporting records within our registered RSP scope. We are not a registered tax agent: each company self-assesses and remains responsible for its own claim, with tax advice and lodgement handled by its tax adviser. Get in touch.
This article is general information from a Registered Research Service Provider about the R&D Tax Incentive. It is not tax, legal or financial advice; eligibility depends on your circumstances and you should self-assess and seek your own advice.
Thinking about a project like this?
If you're weighing up an AI, software or technical improvement project and can't tell yet whether it's implementation or research, start with a quick read on where it sits.