TLDR: Scientific uncertainty in a clinical R&D claim means a competent professional with access to worldwide published knowledge could not know the outcome of your experiment in advance, per s 355-25(1)(a) ITAA 1997. For health-tech and medtech companies the unknown is usually the effect on a pre-specified clinical or behavioural endpoint, not the code.
Describe the gap in the literature, the endpoint, the population and the measurement method, then register with AusIndustry within 10 months of year end (30 April for a 30 June year end).
What scientific uncertainty means under Division 355
It means an outcome that cannot be determined ahead of time on the basis of current knowledge, information or experience, and can only be resolved by a systematic progression of work from hypothesis to experiment, observation and evaluation. That is the language of s 355-25(1) of the ITAA 1997 (Division 355), and it is the language AusIndustry assesses against.
Note what the section does not say. It says nothing about difficulty, cost, novelty of technology, or how long the work took.
The test is epistemic: was the answer knowable, or did you have to run the experiment to find out? A clinical question can be genuinely unknowable even when every component used to answer it is off the shelf.
Clinical uncertainty versus technical uncertainty
Both satisfy the same statutory test, but they sit in different knowledge bases and are assessed against different literature. Keep them in separate core activities, because the competent professional for each is a different person.
| Technical uncertainty | Scientific or clinical uncertainty | |
|---|---|---|
| The unknown | Whether an architecture, model or method will perform at required scale or accuracy | Whether an intervention changes a physiological, symptomatic or behavioural endpoint |
| Competent professional | Senior engineer or ML researcher in the field | Clinician, epidemiologist or health researcher in the therapeutic area |
| Knowledge base | Papers, benchmarks, open-source implementations, standards | Published trials, systematic reviews, registries, regulatory guidance |
| Typical evidence | Benchmark runs, ablation results, load-test data | Protocol, endpoint data, statistical analysis plan, ethics correspondence |
A digital therapeutics company often has one of each: a technical stream on, say, signal processing from wearable data, and a clinical stream on whether the resulting coaching changes a symptom score. Mixing them into a single activity makes both harder to defend.
For whether the trial itself is eligible, see our guide to clinical trial R&D Tax Incentive eligibility. For the engineering-led equivalent of this article, see how to write the technical narrative.
The registration fields where uncertainty is assessed
Uncertainty is not assessed as one paragraph. The AusIndustry registration (online customer portal, current form version released 15 August 2025) splits each core activity into separate fields, each with its own character minimum, and each testing a different limb of s 355-25.
- Sources investigated. What you searched: databases, trials, reviews, guidelines, patents, expert advice.
- Why a competent professional could not know the outcome. The gap those sources left open.
- Hypothesis. The specific, testable proposition.
- Experiment. Design, arms, variables held constant, endpoint, analysis method.
- Observation. What you measured.
- Evaluation. How you assessed the results against the hypothesis.
- Conclusions and new knowledge. What the results established, including negative results.
Writing one narrative and pasting it across all seven is the most common drafting failure. Field two is where the uncertainty lives, and it should not simply restate field one. Our walkthrough of the registration form fields covers the structure in more detail.
The competent professional test in a clinical context
The benchmark is a competent professional in the relevant field with access to knowledge that is publicly available or reasonably accessible anywhere in the world. Not your team, not your advisers, not the Australian market. "New to the company" has never been sufficient.
For a clinical stream, that professional is someone who reads the therapeutic-area literature. So the question you must answer is narrower than it first appears.
What the professional would already know
Assume they know the general mechanism, the existing evidence for in-person or pharmacological versions of the intervention, and the published effect sizes for comparable digital programs. Claiming uncertainty about something a Cochrane review settled will not survive a review.
Where the residual gap usually sits
Genuine clinical gaps in health tech tend to be about transfer, dose and population: whether an effect demonstrated under clinician supervision persists when delivered asynchronously, what engagement threshold is needed to produce it, or whether it holds in a population the published trials excluded.
Moreton Resources Ltd v ISA [2019] FCAFC 120 supports this reading: applying an existing technology in a new context can be a core activity where the outcome could not be known in advance and the purpose is generating new knowledge. But Coal of Queensland [2021] FCAFC 54 is the counterweight. Novelty of context alone is not enough if there is no genuine unknown and no documented systematic progression.
Phrasing a clinical hypothesis AusIndustry can assess
A hypothesis must be scientific, specific, and capable of being validated or invalidated. In GQHC [2024] AATA 409 the tribunal rejected hypotheses that were vague or commercial in nature, and held that observation and evaluation must be evidenced rather than asserted.
A usable clinical hypothesis names five things:
- The population (adults aged 18 to 65 meeting Rome IV criteria for IBS)
- The intervention and dose (an eight-week app-delivered gut-directed hypnotherapy program, minimum five sessions per week)
- The comparator (waitlist control, or an active symptom-tracking arm)
- The endpoint and instrument (change in IBS-SSS at week eight)
- The direction and threshold (a reduction of at least 50 points, the accepted minimum clinically important difference)
If any of those five is missing, the assessor cannot tell what result would have falsified the proposition. That is usually fatal.
Anatomy of a defensible uncertainty statement
Work through four moves, in order, inside the "why a competent professional could not know the outcome" field.
- State what the literature establishes. Cite the settled ground so the gap is visible against it.
- Name the boundary. Identify the precise variable the published evidence does not cover: delivery mode, population, dose, duration, or endpoint.
- Explain why extrapolation fails. Give the mechanistic or methodological reason the known result cannot simply be assumed to carry across.
- Bind it to the outcome you could not predict. Say plainly that the direction and magnitude of the effect could only be determined by measurement.
A statement built this way reads as an argument rather than an assertion, and it maps cleanly onto the sources-investigated field that precedes it.
Wording that weakens a clinical uncertainty claim
Certain phrases get challenged almost every time, because they describe a commercial goal or a development task rather than an unknown outcome.
| Weak wording | Why it fails | Rewrite toward |
|---|---|---|
| "We wanted to build a best-in-class product" | Commercial aim, not a hypothesis (GQHC [2024] AATA 409) | A stated endpoint, direction and threshold |
| "No one had built this before" | Novelty of product, not uncertainty of outcome | The unresolved question in the literature |
| "It was new to our team" | Fails the worldwide-knowledge benchmark | Sources searched and the gap they left |
| "We tried several approaches until one worked" | Trial and error without isolable variables | Variables varied, held constant, and measured |
| "We validated that users liked the feature" | Preference measurement, not a clinical endpoint | A pre-specified clinical or behavioural measure |
| "We optimised onboarding to lift activation" | Falls within the s 355-25(2)(a) market-development exclusion | A separate activity, or exclude it |
Before and after: three rewrites
Each pair below shows a real-world style description and a version rewritten to name the endpoint, comparator, threshold and literature gap. The before versions all fail the same way: they describe what was built, not what was unknown.
Digital therapeutic for IBS
Before: "We developed an app to help IBS sufferers manage symptoms and tested whether users found it helpful."
After: "Published trials support clinician-delivered gut-directed hypnotherapy for IBS, but the evidence base does not establish whether the effect persists when sessions are delivered asynchronously without clinician contact. We hypothesised that an eight-week self-directed program would produce a mean IBS-SSS reduction of at least 50 points at week eight relative to a symptom-tracking control. Direction and magnitude could not be predicted from the supervised-delivery literature."
Remote monitoring in cardiology
Before: "We built an algorithm to detect deterioration earlier than existing tools."
After: "Published early-warning scores are validated on inpatient observation intervals. Whether the same physiological signals, sampled intermittently from a consumer wearable in an unsupervised home setting, retain sufficient sensitivity to flag deterioration at a clinically actionable lead time was not determinable from that literature, given differences in sampling frequency and artefact profile."
Behavioural adherence in chronic disease
Before: "We ran experiments on notification timing to improve engagement."
After: "We hypothesised that adaptive dosing reminders timed to individual circadian patterns would increase medication adherence, measured as proportion of days covered over 12 weeks, by at least 10 percentage points against fixed-time reminders. Adherence was the pre-specified endpoint; app engagement metrics were not used to evaluate the hypothesis."
Note what changed in each: an endpoint, a threshold, a comparator, and a reason the answer was not already in the literature.
Is a clinical endpoint study market research?
No, provided the endpoint is clinical or behavioural and specified before the study runs. The s 355-25(2)(a) exclusion covers market research, market testing, market development and sales promotion. The discriminator is the endpoint and the pre-stated hypothesis, not the tooling.
Running the study on the same A/B infrastructure you use for growth experiments does not trigger the exclusion. Measuring the same intervention on conversion, churn, retention or revenue does.
In practice, if your experiment platform reports both a symptom score and a subscription-retention metric, make clear in the evaluation field which one tested the hypothesis. Where a study genuinely serves both purposes, register the clinical endpoint and leave the commercial analysis out of the activity entirely.
Evidence that backs the uncertainty you describe
The uncertainty statement is only as strong as the records created at the time. The ATO's record-keeping guidance and the case law both point the same way: records created at the time count, records reconstructed later do not.
| Field it supports | Evidence | When it must exist |
|---|---|---|
| Sources investigated | Dated literature and patent searches, clinical advisory notes | Before the activity starts |
| Outcome unknowable | Expert correspondence stating current knowledge cannot determine the outcome | Before the activity starts |
| Hypothesis | Protocol or study plan stating the hypothesis and endpoint | Before the experiment runs |
| Experiment | Statistical analysis plan, ethics submission, randomisation records | Before or during |
| Observation | Raw endpoint data, exports, timestamped analysis notebooks | During |
| Evaluation and conclusions | Analysis outputs including negative results | During or after each cycle |
In Royal Wins Pty Ltd [2020] AATA 4320, no contemporaneous documentation that a hypothesis was developed or tested meant the activities failed, regardless of the volume of work done. In Absolute Vision Technologies [2022] AATA 2319, documents prepared after the fact were not contemporaneous and did not match the registered activities.
Retain records for at least five years. Our guides to contemporaneous evidence and what evidence AusIndustry accepts go deeper on the artefacts that hold up in a review.
Rand builds each activity from the records that already exist in your systems (protocol drafts, ticket history, analysis commits, time logs), so the uncertainty field is drafted against dated artefacts rather than reconstructed at lodgement.
Where clinical uncertainty ends and routine work begins
The core activity is the unresolved question. Everything built to run the experiment can sit under s 355-30 (Division 355) as a supporting activity, provided it is directly related to the core activity, and, where it falls in an excluded category or produces goods or services, is undertaken for the dominant purpose of supporting that core activity.
That usually includes trial infrastructure, randomisation and blinding tooling, data capture and cleaning, endpoint instrumentation, ethics applications and protocol development. Group these into two to four broad supporting categories per core activity rather than one per ticket. Our core versus supporting explainer covers the dominant-purpose overlay in detail.
What the claim is worth
For a company with aggregated turnover under $20M, the offset is the company tax rate plus 18.5 percentage points, commonly 43.5% for a base-rate entity taxed at 25%, and refundable to the extent it exceeds tax payable. For the full rate table, the minimum spend rules and the intensity premium that applies at $20M turnover or more, see our R&D Tax Incentive eligibility guide.
The 2026-27 Federal Budget proposed lifting the $20M refundable threshold to $50M from 1 July 2028. That is an announced measure, not yet law, and it does not affect current income years.
Drafting checklist before you lodge
- The uncertainty field names a gap in worldwide published knowledge, not a gap in your team's knowledge.
- The hypothesis states population, intervention, comparator, endpoint, instrument and threshold.
- A dated record shows the hypothesis existed before the experiment ran.
- The endpoint is clinical or behavioural, not conversion, churn or revenue.
- Sources investigated lists actual searches and advice, with dates.
- Evaluation describes results against the hypothesis, including negative findings.
- Clinical and technical streams are registered as separate core activities.
- Every claimed hour maps to a named activity through a timesheet or equivalent.
- Registration is lodged within 10 months of year end (30 April for a 30 June year end), per the business.gov.au R&D Tax Incentive guidance.
Clinical R&D is often the strongest claim a health-tech company has and the worst documented, because the unknown lives in a protocol rather than a pull request. If you want the uncertainty drafted against evidence that already exists in your systems and reviewed by people who assess these activities for a living, that is what Rand is built to do.