Quick answer: Encoding a design rule set a competent professional already knows into a parametric model and letting it generate and rank options is generally unlikely to be a core R&D activity on those facts, subject to the activity's own facts and the statutory tests. What may be core is work where the solution space, the objective function or the coupling between objectives could not be determined in advance and could only be resolved by a systematic progression of work. You self-assess.
18 August 2026 — this article describes the current rules. The 2026-27 Federal Budget announced R&DTI changes scheduled to start 1 July 2028; those are not yet law.
A generative design run looks like an experiment. Ten thousand facade variants appear overnight, a Pareto front is plotted, a point on it is chosen. Under the R&D Tax Incentive (R&DTI) none of that is decisive. What matters is whether the outcome could have been determined in advance from current knowledge, information or experience — and where the rules and relationships being searched were already known and documented by the designer, the outcome may be determinable in advance.
I write as a Registered Research Service Provider (RSP000047) at Lot Fourteen in Adelaide. We supply research capability and structure R&D activities; we do not give tax advice, and the rules below belong to the ATO, AusIndustry and the legislation, not to us.
Scope note: this article is about generating and searching design options — the encoding, the objective formulation and the search itself. It does not cover automating structural, thermal or FEA analysis, or surrogate models that predict a solver's answer — the subject of our separate Insight on automating structural and engineering analysis and of our property and construction coverage. That article asks whether the analysis engine's behaviour was unknown; this one asks whether the design space and the objectives were.
Writing Down a Rule You Already Know Is Not Generating New Knowledge
Most parametric work translates professional knowledge into machine-readable form. A designer knows the facade module, the setback rules, the daylight rule of thumb, the fabricator's tolerance, the structural grid; a script encodes those relationships and regenerates geometry faster than a person could. The relationships were known before the script existed. The script does not discover them.
The same holds when a search is bolted on. Sweeping a nine-parameter definition with a genetic algorithm produces options nobody had drawn, but "nobody had drawn it" is not the statutory unknown. Where a competent professional could have said in advance roughly where the good region sits, the exercise enumerates a known space, and enumeration at scale is a labour saving rather than new knowledge.
That is why "we used an algorithm" carries little weight on its own. AusIndustry's AI-related activities sub-guide to the software development sector guide puts it directly: "Using an AI model or technique that is new to you does not, by itself, mean the activity is eligible for the program." The same guidance treats parameter tuning by established methods, where the effect of the change is well understood, as usually not core — a fair description of adjusting a population size or mutation rate until a solver behaves.
The Statutory Test, Stated in Full
Section 355-25(1) of the Income Tax Assessment Act 1997 defines core R&D activities as 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 (i) is based on principles of established science and (ii) 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 (legislation.gov.au; business.gov.au).
That is the test as the Act states it. The phrase "competent professional" does not appear in s 355-25(1); it comes from AusIndustry guidance, which frames whether an outcome can be determined in advance by reference to what a competent professional in the field could have known.
Two limbs do most of the work in generative design, and both are routinely skipped:
Established science
Running many configurations and keeping whichever scored best does not, by itself, demonstrate a systematic progression based on principles of established science. What grounds the work here is the building physics, structural mechanics or optimisation theory that says why a given encoding should be able to represent the phenomenon, and which regime boundaries the trials ought to probe. Without that, a generative run is empirical sifting, however many generations it survives.
Logical conclusions
A progression has to conclude something. "Option 4,812 scored best" is a selection. "The weighted-sum formulation cannot represent this trade-off, because the objectives are not exchangeable at a fixed rate across the range of interest" is a conclusion: it rules something out and directs the next trial.
Where the Unknown Can Genuinely Sit
Three things in a generative workflow can be undetermined in advance in a way the rest of it is not:
The solution space: Whether the feasible region is even connected, and whether an encoding can reach it, may be unknown where a hard constraint — fabrication repetition, a discrete component catalogue, a modular grid — interacts with continuous geometry in a way published methods do not cover.
The objective function: Where the quantity being maximised has no accepted formulation for the case at hand, whether any computable objective correlates with the physical behaviour of interest may itself be open.
The coupling between objectives: Two objectives with well-understood individual behaviour can couple non-monotonically through a shared geometric variable, and the shape of that interaction may have no predictive basis for the system in question.
Two Exclusions That Catch a Lot of Design Automation
Subsection 355-25(2) lists activities that cannot be core R&D activities. Two land squarely on this work:
Paragraph (g) — Reproduction
The exclusion covers "any activity related to the reproduction of a commercial product or process: (i) by a physical examination of an existing system; or (ii) from plans, blueprints, detailed specifications or publicly available information". Parametric modelling may raise this exclusion where the activity is actually directed to reproducing a commercial product or process from technical literature, detailed specifications or other specified sources. The output geometry is new; the product or process being reproduced is not.
Paragraph (f) — Statutory requirements and standards
The exclusion covers "activities associated with complying with statutory requirements or standards", including "routine testing and analysis of materials, components, products, processes, soils, atmospheres and other things". Compliance-driven design automation may fall within this exclusion where the activity is undertaken to meet a statutory requirement or standard, including a requirement or standard contained in legislation or imposed by a regulator under legislation. Such work is often sophisticated. It is also, on those facts, generally unlikely to be a core R&D activity, subject to the activity's own facts and the statutory tests: the requirement is published and the criterion fixed.
Neither exclusion puts the work outside the program. Under s 355-30, supporting R&D activities are activities directly related to core R&D activities; but where an activity (a) is of a kind referred to in s 355-25(2), or (b) produces goods or services, or (c) 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. Where a particular design-production or 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.
A Worked Example — Hypothetical and Illustrative Only
The following scenario is entirely hypothetical. It illustrates where a technical boundary falls; it is not a statement that any of this work would be eligible, and eligibility would be self-assessed by the company against the statutory tests.
An Adelaide architecture and engineering practice designs folded metal shading panels for mid-rise facades, fabricated from a catalogue of types where every additional type adds tooling cost.
Baseline and target, fixed in February before any trial: The current method — a designer hand-builds about twelve options and runs each through the practice's validated annual daylight and thermal-load engines — produced, across the last three projects: useful daylight illuminance on 68% of occupied floor area, annual cooling load 84 kWh/m², and 41 panel types after rationalisation. The target: a search formulation reaching at least 78% useful daylight illuminance, cooling load at or below 76 kWh/m², and no more than 12 panel types, in at least 8 of 10 independent runs from different random seeds, within a 20-hour compute budget. Failure was defined in advance as not reaching that band.
Held constant & varied: Held constant across every trial: the daylight and thermal engines, the Adelaide weather file, the occupancy schedule, the glazing specification, the structural grid and floor plate, and three test buildings (sites A, B and C). Varied: the design representation, the objective formulation and the search strategy — nothing else.
Why the outcome was not determinable in advance: Published relationships for shading depth against daylight and cooling load cover continuous flat shading on rectangular facades. This panel family folds in two planes and carries a hard repetition constraint. Whether the two objectives couple monotonically through the fold geometry under that constraint, and whether the feasible region stays connected, had no predictive basis in existing knowledge.
Trial 1 — weighted-sum single objective, nine continuous parameters (failed): The formulation reached 71–74% daylight and 81 kWh/m², and only in 3 of 10 seeded runs; a sensitivity sweep across weight vectors flipped the ranking of the leading options. This ruled out a scalar weighted sum for this problem — the objectives are not exchangeable at a fixed rate over the range of interest, so no fixed weight vector represents the design intent — and with it the assumption that the trade-off surface was convex in that region.
Trial 2 — multi-objective search, same continuous schema, repetition handled afterwards: The front reached 79% daylight at 74 kWh/m². After rationalising to 12 panel types, realised performance fell to 70% and 80 kWh/m², the fall varying between 3 and 9 daylight points depending on which front point was rationalised. This ruled out treating fabrication repetition as post-processing: the constraint is not separable from the geometry search.
Trial 3 — catalogue as a decision variable: The schema was rebuilt so that a fixed-size catalogue of fold types and a per-bay assignment were themselves searched, with the repetition count as a hard constraint. On sites A and B, 8 of 10 runs reached at least 78% daylight at or below 76 kWh/m² with 11–12 panel types, inside 17 hours. On site C — heavily overshadowed by an adjacent tower — only 4 of 10 runs reached the band, the failures clustering where the daylight objective became insensitive to fold geometry.
Result actually reached: On facades not dominated by external overshadowing, the catalogue-as-variable encoding reaches the band repeatably within budget. Where an adjacent building dominates the daylight term it does not, and the year closed without establishing whether that is a search-budget limit or a limit of the encoding — a partial answer and a live open question.
Where the boundary falls: Of roughly 1,450 hours: about 300 sit in the hypothesis, the three trials and their evaluation; about 180 in generating and freezing the baseline option sets and reference runs, which exist only to feed the trials, so whether they attach turns on s 355-30 including dominant purpose; about 640 in the parametric definition, the fabricator export, the interface and documentation, which apply known methods to known outcomes; and about 330 in producing the shading design for the live project and verifying it against applicable National Construction Code energy-efficiency provisions — to the extent that the activity is undertaken to comply with statutory requirements or standards, it falls within the s 355-25(2)(f) exclusion from core R&D. Had the catalogue been rebuilt from the fabricator's published specification, s 355-25(2)(g) would also be in play. How each stream is treated in a registration and in the company's tax return is for the company and its registered tax agent to determine.
Where an RSP Fits
AusIndustry describes Research Service Providers as organisations with the scientific or technical capability to conduct R&D on your behalf, registered in specific research fields (business.gov.au). Its guidance notes that AI-related activities "may meet the requirements of a core R&D activity where a technical hurdle exists and an expert in the field considers that only experimentation will determine if a proposed solution, or the way to develop a solution, can resolve the technical hurdle" — a question relevant to assessing whether the core R&D requirements are met. An RSP may assist with planning and conducting R&D activities within the research fields for which it is registered.
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 are within a research field for which the RSP is registered (business.gov.au) — see claiming R&D under $20,000. Using an RSP does not guarantee eligibility — you still self-assess. Offset rates, the refundable and non-refundable tiers and the entitlement rules are covered separately in our refundable vs non-refundable offset article.
The general version of that line sits in our software and AI and what does not qualify pages.
Frequently Asked Questions
Q: Is parametric design eligible for the R&D Tax Incentive?
A: Encoding known design rules, module logic and clearances into a parametric definition so geometry regenerates automatically is generally unlikely to be a core R&D activity on those facts, subject to the activity's own facts and the statutory tests — the relationships were known before the script was written. The separate question is whether any part of the work turned on a solution space, objective or coupling that could not be determined in advance from current knowledge, information or experience. You self-assess.
Q: Is a generative design algorithm a core R&D activity?
A: Running a search is not, by itself, an experiment for R&DTI purposes. Section 355-25(1) requires an outcome that could not be determined in advance, a systematic progression of work based on principles of established science that proceeds from hypothesis to experiment, observation and evaluation and leads to logical conclusions, and the purpose of generating new knowledge. A generative run that enumerates a space whose shape a competent professional could describe in advance does not meet that.
Q: Is automating code-compliant design generation eligible R&D?
A: Where the activity is undertaken to comply with a statutory requirement or standard, it cannot be a core R&D activity under s 355-25(2)(f). Such work may still qualify as a supporting R&D activity where it is directly related to a core activity and, where the additional test in s 355-30 applies, is undertaken for the dominant purpose of supporting that core activity.
Q: Does rebuilding a facade system from a manufacturer's specification count as R&D?
A: Section 355-25(2)(g) excludes any activity related to the reproduction of a commercial product or process by physical examination of an existing system, or from plans, blueprints, detailed specifications or publicly available information. Rebuilding a supplier's module logic parametrically from technical literature may fall within that exclusion where the activity is directed to reproducing a commercial product or process from the specified sources.
Sources & Further Reading
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25 and 355-30
ATO — Tax Reform: better targeting the R&D Tax Incentive — the 2026-27 Federal Budget announcement of 12 May 2026, the proposed changes, and the ATO's statement that the measure is not yet law
industry.gov.au — Research and Development Tax Incentive — the proposed commencement: changes will apply to income years starting on or after 1 July 2028
Related: property and construction · software and AI · what does not qualify · what an RSP is · claiming R&D under $20,000 · refundable vs non-refundable offset · R&D Tax Incentive in Adelaide
Talk to Ignition Research before you register a year of generative and parametric design work — as a Registered Research Service Provider at Lot Fourteen in Adelaide, we work with architecture, engineering and design-software teams on what is genuinely unknown about a design space, what the objective is measuring, and what a progression from hypothesis to logical conclusions looks like in this discipline. We are not a registered tax agent: your company self-assesses and remains responsible for its own claim, with advice and lodgement handled by your 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.