Quick answer: Modelling how a business runs in order to find savings sits in the territory s 355-25(2)(c) names — management studies or efficiency surveys — and is generally unlikely to be a core R&D activity on those facts, subject to the activity's own facts and the statutory tests. A core R&D activity may sit narrower: where it is unknown whether any formulation or algorithm produces a usable solution inside the operational constraints, determinable only by a systematic progression of work. You self-assess.
18 August 2026 — this article describes the current rules. The 2026-27 Federal Budget announced proposed changes to the R&DTI; the ATO states the measure is not yet law, and industry.gov.au states the changes would apply to income years starting on or after 1 July 2028 if enacted.
An optimisation project usually starts with a number — changeover hours, empty kilometres, the overtime the roster keeps producing. That number is a business case, not a statutory test, and the size of the saving says nothing about whether any of the work is a core R&D activity. Division 355 does more than ignore the saving: one paragraph of the exclusions names work of this shape directly.
This article is about the formulation and the solve in scheduling, routing and supply-chain optimisation. It does not cover AI quoting and estimating engines, or forecasting and decision-support systems that recommend a decision to a person; both are treated separately in our Insights.
The Exclusion That Sits Directly on This Work
Section 355-25(2) lists activities that cannot be core R&D activities, however carefully conducted. Paragraph (c) is short:
"management studies or efficiency surveys"
The Act does not define either term. What it does is remove the activity from the core category by its character, not by its rigour — a study can be quantitative, statistically designed and genuinely useful and still be an efficiency survey. Mapping the current process, measuring changeover and travel times, and quantifying cost or time savings may fall within this exclusion where the activity collects and evaluates information to inform decisions about business operations or their efficiency. Whether a particular simulation or optimisation activity falls within the exclusion depends on its character and purpose. Whether a particular activity is a management study or an efficiency survey is a question of fact, assessed activity by activity, and the company self-assesses. An activity caught by s 355-25(2) can still be a supporting R&D activity, but only on the stricter condition set out below.
The Core Test, Stated in Full
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 (s 355-25(1), ITAA 1997; business.gov.au).
Two limbs do most of the work here:
The unknown outcome limb
The unknown outcome limb has to bite on something technical. That the schedule can be improved is usually known — an operations manager with a spreadsheet can demonstrate it. The unknown, if there is one, is whether a formulation or algorithm exists that returns a usable solution under the real constraints: the full instance size, the hard operational rule that cannot be relaxed, and the wall-clock budget the operation allows.
The established science limb
Operations research is established science — integer and constraint programming, complexity, relaxation strength, duality, queueing and graph theory — but the limb asks that the progression of work rest on those principles, not merely that they exist in the field. Changing solver settings until a number improves rests on trial and error; predicting and then measuring a bound, isolating why a relaxation is weak, and ruling a family of formulations in or out on that evidence rests on established science.
When It Is Configuring a Solver
AusIndustry's guidance on AI-related activities lists work that is generally not a core R&D activity. It is written for AI, but the reasoning is the one s 355-25(1) applies, and two items describe optimisation projects closely, verbatim:
• "adjusting or fine-tuning parameters (e.g. learning rate) using established methods when it is known that the issue can be solved by parameter changes and the effect of changing the parameter is well understood"
• "running regression, acceptance, or functionality tests using established testing methods to confirm a system works as intended, where expected outcomes are already known"
The same guidance is direct about novelty of tooling: "Using an AI model or technique that is new to you does not, by itself, mean the activity is eligible for the program" (business.gov.au). A first encounter with a commercial mixed-integer solver, or with a routing library, is a first encounter.
On that reasoning, expressing a vehicle-routing problem with time windows in a documented library, selecting between solvers on published benchmarks, tuning search parameters where the effect is understood, adding capacity to the solve server, re-running the model on new demand data, and building the integration around it are generally unlikely to be core R&D activities on those facts, subject to each activity's own facts and the statutory tests. See what does not qualify.
When the Outcome Is Not Knowable in Advance
The guidance states the other side too: "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."
For an optimisation engine that hurdle is usually structural. Signals that recur:
The instance defeats the standard model at scale: A formulation that closes to a small gap on a reduced instance leaves a wide bound at full size, and compute barely moves it — the failure is in the relaxation, not throughput.
The published results are on different structure: Benchmark instances carry particular distributions; a plant with sequence-dependent cleans, or a network with a regulated rest rule, is not that instance.
The constraint is not in the standard model: A rule the operation cannot break, expressed as a transition, a disjunction or a coupling across resources, changes what any relaxation can bound.
There is a re-solve budget: An answer that arrives after the decision is not a solution, and whether any formulation holds a seconds-scale budget on disrupted instances may be unknown.
None of this is established by asserting difficulty. The activity still needs a testable hypothesis and an appropriate evaluation approach within the required systematic progression of work.
Other Exclusions Across an Optimisation Project
Three further paragraphs of s 355-25(2) recur here:
Paragraph (e) — Commercial, legal and administrative aspects
Commercial, legal and administrative aspects of patenting, licensing or other activities — the solver licence negotiation, not the mathematics.
Paragraph (f) — Statutory requirements and standards
Activities associated with complying with statutory requirements or standards, "including one or more of the following": (i) maintaining national standards; (ii) calibrating secondary standards; (iii) routine testing and analysis of materials, components, products, processes, soils, atmospheres and other things. Routing and rostering work may touch statutory or regulatory requirements — for example, applicable heavy-vehicle fatigue requirements or regulated food-handling requirements. "Associated with" is broad, and whether development work aimed at meeting a standard falls inside paragraph (f) is a question of fact, assessed activity by activity; the company self-assesses.
Paragraph (g) — Reproduction of a commercial product or process
Any activity related to the reproduction of a commercial product or process by a physical examination of an existing system, or from plans, blueprints, detailed specifications or publicly available information. Both elements are required: reimplementing a published metaheuristic from its paper and reference code goes to the second element, but a published method or open-source code is not by itself a commercial product or process.
Supporting R&D activities are activities directly related to core R&D activities; however, where an activity (a) is an activity 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 (business.gov.au). Each limb is tested against the particular activity, never against the project or the business as a whole. In optimisation work all three turn up together: a time study building the changeover matrix may engage limb (a); running a candidate schedule on the live line may produce, or be directly related to producing, goods; and a plant data extract that also feeds the production planner may likewise be directly related to producing goods or services. Each activity must be assessed on its own facts.
A Worked Hypothetical: A 90-Second Re-Solve on a Plant That Will Not Sequence
Hypothetical and illustrative. The figures are invented to show the shape of an experiment; nothing here indicates that any activity is eligible.
An Adelaide contract manufacturer packs chilled ready meals on three lines, roughly 140 jobs a week. Changeovers are sequence-dependent, and one rule is hard: a job containing a declared allergen must be followed by a validated wet clean unless the next job carries the same allergen. Lines stop unpredictably, and the supervisor then needs a revised plan before the next changeover window.
Baseline: The current rule — longest-processing-time dispatch with manual allergen clustering — produces 19.4 hours of changeover time per week across the three lines.
Requirement, recorded 3 March: A method is accepted only if it returns a feasible full-week schedule with total changeover time at or below 14.0 hours and zero violations of the allergen-clean rule, and returns a revised schedule within 90 seconds of wall clock on the existing eight-core server across a fixed replay set of 40 recorded mid-shift disruptions. Miss either and the project stops.
The search, recorded the same week: Benchmark results for parallel-machine scheduling with sequence-dependent setups, the two licensed commercial solvers, and the plant's MES logs. Published instances carried setup matrices without a forced-clean transition and reported solve times in minutes, not seconds. Hypothesis: the allergen rule can be carried as a sequence transition and the instance decomposed so that a feasible schedule inside the changeover target is reachable within the budget. Basis: integer programming duality and relaxation strength, and set-partitioning decomposition with column generation.
Held constant & varied: Held constant across trials: the 140-job instance, the setup-time matrix, the replay set of 40 disruptions, the server, and the acceptance measures. Varied: only the formulation and the solve strategy.
Trials:
Trial 1 — Time-indexed MILP with big-M sequencing constraints (Failed): On a 40-job instance it closed in 11 s. At full size, 90 s left a 34% gap and a worse schedule. Raising to 600 s moved the gap only to 29%. Profiling showed node throughput was not binding — the relaxation was weak because big-M constraints gave almost no bound once cleans were forced. This showed that simply extending solve time did not overcome the weak relaxation.
Trial 2 — Constraint programming with interval variables and transition constraints: Feasible schedules at 15.8 hours within budget. The allergen rule propagated cleanly as a transition; target was missed, but the encoding was confirmed to work.
Trial 3 — Set-partitioning decomposition (column generation over the week): 13.6 hours, median solve 71 seconds — but on the 40-disruption replay set, 6 runs exceeded 90 seconds (worst case 148 seconds). Time budget was not met.
Trial 4 — Same decomposition + bounded label-setting routine + warm-started column pool: Median 38 seconds, worst case 84 seconds over the replay set; 13.9 hours of changeover; zero allergen-rule violations. Both conditions met.
The result reached: On this plant's structure, a set-partitioning decomposition with bounded pricing held the hard constraint inside a seconds-scale budget where a monolithic time-indexed MILP could not. The evidence pointed to formulation and relaxation strength under the forced-clean transition as the main constraint in the tested monolithic MILP, rather than compute time.
The activity boundary: For this example, the candidate experimental activity is documented from the formulation of the hypothesis and experimental approach through the trials, replay evaluation and recorded conclusion. The MES extract, the time study that produced the setup matrix, the supervisor interface, integration with the planning system, and running the schedule week to week afterwards are separate activities, each tested on its own facts. Describing work this way is not classifying it: whether any of it meets s 355-25 or s 355-30 is determined on the facts of each activity, and the company self-assesses.
Where an RSP Fits
A Research Service Provider is a scientific or technical service provider, registered in specific fields, that a company can engage to conduct R&D activities on its behalf (business.gov.au). There is also a threshold point for smaller claimants: 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, and using an RSP does not guarantee eligibility — you still self-assess (business.gov.au). See claiming R&D under $20,000; which offset applies, and on what conditions, is covered separately in refundable vs non-refundable offset.
Talk to Ignition Research before you register an optimisation project — as a Registered Research Service Provider at Lot Fourteen in Adelaide, we work upstream of any claim: setting the acceptance measures and the time budget before the work, and designing the trial sequence so the formulation question is answered on evidence. We are not a registered tax agent; your company self-assesses and remains responsible for its own claim. Get in touch.
Frequently Asked Questions
Q: Are efficiency studies excluded from the R&D Tax Incentive?
A: Section 355-25(2)(c) provides that "management studies or efficiency surveys" cannot be core R&D activities. That does not remove them from the program entirely: an activity of a kind referred to in s 355-25(2) may still qualify as a supporting R&D activity where it is directly related to a core R&D activity and undertaken for the dominant purpose of supporting that core activity. Whether a particular activity is a management study or efficiency survey is a question of fact, and you self-assess.
Q: Is route optimisation software eligible R&D in Australia?
A: Not by category. Expressing a routing problem with time windows in a documented library and tuning it with established methods is generally unlikely to be a core R&D activity on those facts, subject to the activity's own facts and the statutory tests. A core R&D activity may exist where it could not be determined in advance, on current knowledge, information or experience, whether any formulation returns a usable solution under the real constraints and time budget.
Q: Is configuring a solver a core R&D activity?
A: Generally not on those facts. AusIndustry's guidance treats "adjusting or fine-tuning parameters ... using established methods when it is known that the issue can be solved by parameter changes and the effect of changing the parameter is well understood" as work that is generally not a core R&D activity. The core test requires an outcome that cannot be known or determined in advance and a systematic progression of work based on principles of established science that leads to logical conclusions.
Q: When does supply chain optimisation become R&D?
A: When the unknown stops being commercial and becomes technical. Quantifying a saving from a process change is efficiency-survey territory. The position can differ where a hard operational or regulatory constraint, the instance size and a fixed re-solve budget together mean that no known formulation or algorithm is known to produce a usable solution, and only experiment can determine whether one does — and only the required systematic progression of work can determine whether a usable solution can be achieved.
Sources & Further Reading
business.gov.au — AI-related activities and the R&D Tax Incentive
ATO — Tax Reform: better targeting the R&D Tax Incentive — the announced measures and the statement that the measure is not yet law
industry.gov.au — Research and Development Tax Incentive — the proposed changes would apply to income years starting on or after 1 July 2028 if enacted
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25 and 355-30
Related: R&D for software and AI · what does not qualify · what an RSP is · claiming R&D under $20,000 · refundable vs non-refundable offset · Insights
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.