Quick answer: Buying AR headsets, deploying a commercial digital twin platform, authoring model content and training crews is generally implementation rather than core R&D — and where the exercise is really about workflow efficiency or staff capability it sits close to the excluded categories in s 355-25(2) of the ITAA 1997. A core R&D activity may sit where the achievable technical outcome was genuinely unknown: whether spatial registration can hold a stated tolerance against a partially built, changing structure; whether tracking stays robust through dust, glare and occlusion; whether a twin can stay synchronised at a usable latency. An unknown target is not enough on its own — the activity must also apply a systematic progression of work and be conducted for the purpose of generating new knowledge. Work done specifically for such an experiment, such as capturing a controlled reference scan, may qualify as a supporting activity. You self-assess.
7 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.
An AR set-out workflow and a live digital twin are new to almost every builder that deploys them, but neither novelty nor difficulty is the statutory test. The R&D Tax Incentive asks whether the outcome of an activity could have been known or determined in advance on the basis of current knowledge, information or experience; whether it could 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 whether it was conducted for the purpose of generating new knowledge (business.gov.au).
The established-science limb is the one immersive work most often misses: a disciplined loop of changing a setting, measuring and changing it again is not a systematic progression unless it rests on principles of established science — survey and geodetic error theory, optics, photogrammetry, sensor fusion — and runs from a stated hypothesis through experiment, observation and evaluation to a logical conclusion. On an immersive programme the part of the spend that may answer all of that is a small, specific set of measurement problems.
Most of an Immersive Rollout Is Implementation
Ordinarily not core R&D activities: deploying a commercial AR platform (headsets, tablets, licences, device management, site network coverage); authoring content (model preparation, level-of-detail reduction, federating trade models, 4D sequencing, placing markers and control points); integrating through documented APIs; standing up dashboards, clash reports, defect registers and as-built records; and training crews and changing the process.
Efficiency and market testing risks
Subsection 355-25(2) of the ITAA 1997 lists categories that cannot be core R&D activities at all, and two sit close to an immersive rollout: management studies or efficiency surveys, and market research, market testing or market development, or sales promotion. A structured trial measuring whether AR instructions reduce rework hours or lift crew productivity is in substance an efficiency exercise; a demonstration built to win the next tender is closer to market development. A headset does not change either characterisation.
VR induction and training modules usually fail the core-activity test on their own facts rather than by exclusion: the outcome of building the module is known before it starts, and the work is not conducted to generate new knowledge. Whether crews adopt it is a genuine question, but an adoption study investigates behaviour and workflow. Experiments necessarily involve observation — what is missing here is the experimental logic (an unknown technical outcome, a measure fixed in advance, trials that could have failed) and the new-knowledge purpose. The general treatment of ineligible work sits on our what does not qualify page.
Where a Core Activity Can Sit
Commercial spatial-computing systems may have documented performance assumptions or specifications that do not fully address the conditions encountered on a changing construction site. A live site is none of those. Where whether a stated level of accuracy, robustness or latency is attainable in those conditions genuinely cannot be answered from current knowledge, information and experience, an activity investigating it may be a core activity. The site conditions below are what such an activity actually investigates:
Site condition, and the question it opens
Usually implementation
Feature collapse and as-built divergence: Stripped soffits and screeded floors give feature-based tracking little to hold, and the model is the wrong reference where a set-down was poured low. Whether registration can hold a stated tolerance against a partially built, changing structure.
Aligning to markers or survey control using the vendor's documented procedure
Marker unavailability and drift: Fiducials are occluded by formwork and props and moved by trades; how far cumulative error runs before re-anchoring is rarely predictable from the specification. Whether a drift-correction or sensor-fusion approach holds tolerance across a full shift.
Re-registering more often, or adding control points, to manage drift operationally
Specular and infrared interference: Reflective duct insulation, wet slabs and direct sun through facade openings corrupt depth returns, making time of day a variable. The achievable robustness envelope under dust, glare, low light and occlusion.
Adjusting exposure, improving task lighting, or restricting use to good conditions
Noisy, intermittent, high-volume telemetry: Whether a twin can be held synchronised at a usable latency and fidelity.
Connecting sensors and scanners to a commercial twin through documented APIs
VR at scale: Whether a multi-user session holds interaction latency and state consistency within a stated bound on defined hardware, at a model size no published result covers.
Running a design review in a vendor's VR environment at a supported model size
Two limits on that left column. The unknown is measured against knowledge available in the field, so the search is a dated artefact held against the specific metric — vendor specifications and release notes for the SDK version in use, applicable standards, published literature — and one recorded after the trials reads as reconstruction. And programme pressure and cost exposure are business risks, not indeterminate outcomes. The same boundary in software sits on our R&D for software and AI page.
Supporting Activities and Mixed-Purpose Work
Activities that are not core may still be supporting R&D activities. Under s 355-30 of the ITAA 1997 they must be directly related to core R&D activities — and where the activity is of a kind excluded from being a core activity, produces goods or services, or is directly related to producing goods or services, only where it is conducted for the dominant purpose of supporting a core activity (business.gov.au).
The awkward case is the shared scan. A weekly laser scan is usually captured to produce the progress and deviation reporting the client pays for — production, directly related to producing services — and the same run may also feed a registration experiment. That does not convert the run into supporting work, and describing it differently afterwards does not change the purpose it was conducted for. What separates the two is incremental work done for the experiment: stations set up only to cover the test area, higher point density than reporting requires, a second pass at another time of day to capture a lighting variable, extra control targets observed for ground truth. Those increments exist at the time and are recorded per run; whether an allocation between them is sound, and how the expenditure is treated, is for the company and its registered tax agent.
Subsection 355-25(2) also excludes software activities whose dominant purpose is the internal administration of business functions of the developer, an entity connected with it, or an affiliate. A twin built mainly to run a builder's own project controls needs testing against that limb rather than assuming it sits outside.
A Hypothetical Worked Example
Illustrative and hypothetical only. It is not a ruling, is not based on any client, and the facts below do not establish that anything would be eligible.
An Adelaide contractor is using AR tablets for mechanical and hydraulic penetration set-out on in-progress suspended slabs of a mid-rise building. In the scenario, the platform's documentation quotes ±10 mm registration accuracy at 5 m in a static, finished, well-lit interior, and survey of the first three floor plates found as-built divergence from the model of up to 35 mm.
Target, fixed before any trial: ±20 mm horizontal at up to 10 m from control, on 90% of check points, held for 60 minutes of continuous use. The figure came from the trades' set-out tolerance, went into a one-page test plan, and was not revised afterwards.
Design: Twelve check points per plate, independently observed by total station as ground truth. Held constant — device, SDK version, model version, control network adjustment. Varied one at a time — registration method, time of day, plate. Three methods, four runs each, across three plates, with lux and visible dust logged per run.
Method A — printed fiducials on columns: Failed on availability: six of twelve points unmeasurable on two plates, markers occluded by props and stacked material or moved. Ruled out marker-only registration on in-progress plates.
Method B — unaided feature-based tracking off the as-built environment: Failed differently: on stripped soffits and screeded floors tracking was lost within about four minutes, cumulative error past 60 mm before loss. Ruled out running without an external anchor.
Method C — coarse scan-to-model alignment re-anchored to two surveyed prisms, with drift correction from inertial data: Partially successful: median error 14 mm, 90th percentile 26 mm, target held roughly 90 minutes in the morning. On the western plate after 2 pm, direct sun through facade openings pushed the 90th percentile past 40 mm.
Result: The target was met under stated conditions, not met under direct sun, and the direct-sun case was left unresolved when the programme moved on. Roughly 310 hours sat inside the trials against about 2,100 hours of programme delivery, recorded per run against a separate job code.
Boundary: For this illustrative example, the candidate experimental activity is documented from the point at which the technical hypothesis and evaluation approach are established through to the trials, evaluation and recorded conclusion. Using Method C as the standard set-out procedure on later floors is delivery, not experiment, even though the technique came out of the trials. The vertical picture sits on our R&D for property and construction page.
What the Research Record Looks Like
Immersive programmes generate enormous evidence of delivery — device logs, scan archives, model versions, rollout plans — and very little evidence of inquiry. AusIndustry expects records showing the activities were conducted as registered and how they met the eligibility criteria (business.gov.au). Concretely, on this work:
The activity is defined below programme level: One activity should not combine unrelated technical uncertainties. Registration accuracy on in-progress structure is not the same activity as twin synchronisation latency.
Evaluation measures should be documented contemporaneously: Any material changes during the experimental work should be recorded and technically justified. A tolerance that first appears after the runs describes an outcome, not a hypothesis.
Versioning covers what moves the result: SDK and firmware versions, calibration files, control network adjustments, model version, and a per-run environment log.
Raw per-point residuals are retained: Not only summaries. Failed calibration runs and abandoned methods are the strongest evidence the outcome was unknown, and the first thing lost in a cleanup.
Time is separated as it is worked: Scan-capture and analysis hours reconciling to individual test runs rather than to the BIM or project-controls workstream.
The common failure mode is reconstruction: the programme runs as delivery and the experiment is assembled from delivery artefacts months later, when no variable was held constant and nothing that failed was kept.
Where an RSP Fits
AusIndustry describes Research Service Providers as scientific or technical service providers that can be engaged to conduct R&D activities on a company's behalf, registered in specific fields (business.gov.au). As an RSP at Lot Fourteen, Ignition Research may assist with experimental design, technical R&D work and contemporaneous supporting records within the research fields for which it is registered. Earlier involvement can help document the technical question, hypothesis and evaluation approach as the work progresses. We are not a registered tax agent, and using an RSP does not guarantee eligibility — each company must self-assess and remains responsible for its own claim.
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 — see claiming R&D under $20,000 and what an RSP is. Offset rates and how they apply are covered on our refundable vs non-refundable offset page.
Frequently Asked Questions
Q: Is deploying AR headsets on a construction site eligible for the R&D Tax Incentive?
A: Generally not on its own. Procuring devices, deploying a commercial AR platform, authoring model content and rolling it out to crews is implementation — the outcome is known before it starts. A core R&D activity may exist where a technical outcome such as achievable registration accuracy or tracking robustness in your conditions could not be known in advance, could only be determined by a systematic progression of work, and was pursued to generate new knowledge. You self-assess.
Q: Is building a digital twin R&D?
A: Connecting scanners, sensors and program data to a commercial twin through documented interfaces is integration rather than experimentation. What may be a core activity is establishing whether the twin can be held synchronised at a usable latency and fidelity from noisy, intermittent site data where that was genuinely unknown. A twin built mainly to run a company's own project administration should also be tested against the internal-administration software exclusion in s 355-25(2) of the ITAA 1997.
Q: Is developing AR or VR training content for site crews claimable?
A: Routine authoring of AR or VR training content is unlikely to be a core R&D activity where its technical outcome can be determined in advance and the activity is not conducted to generate new knowledge. A separate study directed at measuring workflow efficiency or employee productivity may also need to be assessed against the management-study or efficiency-survey exclusion. Such activities may still qualify as supporting R&D activities where they are directly related to a core activity and satisfy the relevant supporting activity requirements.
Q: What part of an AR construction workflow can be a core R&D activity?
A: The parts where achievable performance was the unknown — registration and drift correction against a partially built, changing structure; tracking robustness under dust, glare and occlusion; holding a twin synchronised at a usable latency from noisy site telemetry. Each still needs an outcome that could not be known or determined in advance, the required systematic progression of work based on established science, and a purpose of generating new knowledge, assessed on its own facts.
Sources & Further Reading
legislation.gov.au — Income Tax Assessment Act 1997 — Div 355, incl. ss 355-25 and 355-30
Related: property and construction · what does not qualify · software and AI · what an RSP is · under $20,000 · offset rates · more Insights
Ignition Research works with builders, developers and construction-technology teams on the research side of an immersive site programme — defining the activity below programme level, fixing the measure before the trials run, and designing the tests so the unknown is actually resolved. As a Registered Research Service Provider at Lot Fourteen in Adelaide we supply and structure research capability; we are not a registered tax agent, and your company self-assesses and remains responsible for its own claim, with advice and lodgement from 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.