This report examines 8,017 real AI interactions across two Australian NDIS disability providers. Rather than modelling what AI could theoretically do, we show what clinicians chose to use it for, how much time each task type saves, and what share of a therapist's working week AI already covers.
The workforce problem
Australia's disability and aged care workforce grew 70 per cent in five years to 2025, with projected growth more than double the rate expected for all occupations. Despite this, the sector remains in persistent shortage across Australia (Jobs and Skills Australia, 2025). Therapists funded under the NDIS — occupational therapists, physiotherapists, speech pathologists, behaviour support practitioners, nutritionists, and their managers — are central to this shortage.
The nature of NDIS work makes this worse. Every intervention must be planned, evidenced, and reported against a participant's funded goals. Therapists in NDIS disability care carry a documentation burden substantially heavier than their equivalents in private practice or hospital settings. Every hour spent on documentation is an hour not spent delivering care to people who need it.
AI will not solve the workforce shortage by replacing therapists. But if it can absorb enough of the documentation and administrative burden, it can free the therapists you already have to spend more of their time on clinical work. The question is: does it, and for what?
What therapists use AI for
Across 8,017 AI-assisted interactions, therapists use AI for 50 distinct task types spanning clinical documentation, information retrieval, and decision support. The distribution is steep — a small number of high-frequency tasks account for the majority of usage, with a long tail of specialist tasks used less often.
The top tasks by volume are:
- Edit or reformat text (2,420)
- Draft emails to families or providers (579)
- Summarise notes over time (474)
- Look up therapy and medical records (473)
- Look up client profile and demographics (438)
- Write progress reports and reviews (382)
- Summarise therapy or family contact (331)
- Look up assessments (292)
- Look up equipment and assistive tech (279)
- Create notes, e.g. SOAP (278)
- Recommend care interventions (246)
- Write NDIS documentation (171)
- Write assessments (148)
- Look up care requirements (141)
- Look up funding, e.g. NDIS, HCP, AN-ACC (130)
Task-level time savings
Each task type has a distinct time-savings profile. Progress reports save 45 minutes per instance but are written less frequently than quick text edits that save 5 minutes each. The top 20 tasks ranked by total hours recovered:
| Task | Category | Count | Hours saved | Median | Helped |
|---|---|---|---|---|---|
| Edit or reformat text | Writing | 2,420 | 380 | 5 min | 99% |
| Write progress reports and reviews | Writing | 382 | 230 | 40 min | 98% |
| Draft emails to families/providers | Writing | 579 | 159 | 12 min | 97% |
| Summarise notes over time | Info retrieval | 474 | 136 | 15 min | 96% |
| Look up therapy and medical records | Info retrieval | 473 | 105 | 15 min | 93% |
| Write NDIS documentation | Writing | 171 | 97 | 30 min | 98% |
| Create notes (e.g. SOAP) | Writing | 278 | 91 | 20 min | 97% |
| Summarise therapy or family contact | Info retrieval | 331 | 87 | 15 min | 96% |
| Recommend care interventions | Decision support | 246 | 86 | 20 min | 98% |
| Write assessments | Writing | 148 | 83 | 30 min | 96% |
| Write equipment funding justification | Writing | 107 | 66 | 35 min | 98% |
| Look up assessments | Info retrieval | 292 | 54 | 12 min | 86% |
| Look up client profile and demographics | Info retrieval | 438 | 43 | 2 min | 90% |
| Draft clinical handover documents | Writing | 62 | 32 | 30 min | 95% |
| General overview of person | Info retrieval | 107 | 31 | 15 min | 98% |
| Look up equipment and assistive tech | Info retrieval | 279 | 30 | 5 min | 80% |
| Write care plans | Writing | 45 | 25 | 35 min | 100% |
| Look up care requirements | Info retrieval | 141 | 24 | 10 min | 96% |
| Look up communication profile | Info retrieval | 93 | 15 | 10 min | 95% |
| Write behaviour support plans | Writing | 21 | 13 | 25 min | 95% |
The highest per-instance time savings come from complex documentation: progress reports (45 min), NDIS documentation (35 min), clinical handover documents (45 min), and care plans (35 min). These are the tasks where AI replaces the most manual effort in a single sitting. Information retrieval tasks like summarising notes (15 min) and looking up records (10 min) save less per instance but accumulate across every clinical encounter.
What this looks like in practice
The numbers above describe what therapists use AI for. The examples below show what that looks like in a working day.
Progress report — Occupational Therapist
An OT preparing a 6-month NDIS plan review for a participant with complex needs. Mini pulls together the participant's goal history, session notes, and outcome measures, then drafts a structured progress report against their funded goals. The OT reviews it, adjusts clinical language, and submits. What would have taken 45 minutes of manual compilation is done in 10.
Clinical handover — Physiotherapist
A physio transferring a participant to a new therapist in another region. Mini summarises the full treatment history — initial assessment, intervention plan, progress notes, and equipment recommendations — into a handover document the new therapist can use immediately. 45 minutes saved per handover.
Behaviour support plan — BSP
A behaviour support practitioner reviewing incident reports across the last 3 months to update a participant's positive behaviour support plan. Mini identifies patterns across incidents, summarises triggers and outcomes, and drafts updated strategies grounded in the data. The BSP validates the clinical reasoning and finalises.
Session preparation — Speech Pathologist
An SLP preparing for a dysphagia review. Mini retrieves the participant's mealtime management plan, recent nursing notes on swallowing observations, and the last assessment results. The SLP arrives at the session with full context instead of spending 15 minutes pulling records.
How much of the job AI covers
Individual task savings tell one side of the story. The other is: what fraction of a therapist's actual job does AI currently cover? To answer this, observed Minikai usage is mapped onto O*NET occupational task structures, which define every core task in a role and rate how frequently each is performed.
For each O*NET task, coverage is scored 1.0 if usage data shows therapists routinely completing it with AI, 0.5 if AI assists with part of it but not all, and 0.0 if there is no AI usage for it. AI usage is then the frequency-weighted coverage across all tasks in the role.
| Discipline | Tasks | AI coverage | AI usage (time %) |
|---|---|---|---|
| Behaviour Support | 27 | 19 (70%) | 43.9% |
| Therapist Manager | 18 | 11 (61%) | 42.7% |
| Physiotherapist | 24 | 14 (58%) | 37.5% |
| Occupational Therapist | 17 | 10 (59%) | 35.5% |
| Speech Pathologist | 23 | 9 (39%) | 29.7% |
| Nutritionist | 28 | 13 (46%) | 29.3% |
Across the six therapist disciplines, AI covers between 29 per cent and 44 per cent of the working week — meaning close to a third of a therapist's time on formal job tasks is already being augmented. The tasks AI does not touch are clinical judgement, hands-on therapy, and direct patient interaction. The tasks it does cover are documentation, record retrieval, and administrative reporting.
How usage deepens over time
AI adoption is not static. Therapists who find value in AI progressively expand their usage across more of their workflow. The data shows a consistent pattern: clinicians begin with quick, low-risk tasks — editing and reformatting text, drafting emails — and over the following weeks expand to more complex documentation tasks like progress reports, NDIS documentation, clinical handovers, and care plans.
The quick-win tasks (text editing at 5 min median, emails at 12 min) are not where the biggest time savings come from. But they are where clinicians build confidence. Once a therapist trusts the tool on a simple edit, they try a progress report (45 min saved). Once that works, they try NDIS documentation (35 min saved). The total time recovered per clinician grows as their usage expands across task types.
For providers, this means the first month of AI deployment is not representative of long-term value. The highest-impact tasks — the ones that save 30 to 45 minutes per instance — take weeks to appear in usage patterns. Early metrics will be dominated by quick tasks. The time savings compound as clinicians gain fluency.
What this means for providers
AI supports both documentation and clinical decision-making.Across 8,017 task instances, AI usage spans documentation, information retrieval, and clinical decision support. AI drafts progress reports and clinical notes, but it also retrieves and synthesises information from across a participant's history — helping therapists make more informed clinical decisions. The tasks AI does not touch are hands-on therapy and direct participant interaction. AI is not replacing therapists — it is giving them better information and more time to deliver care.
The time savings are real and measurable. 1,940 hours recovered across two providers. At the task level, a single progress report saves 45 minutes, an NDIS plan review saves 35 minutes, and a clinical handover saves 45 minutes. These are not theoretical estimates — they are measured against actual task completion times and calibrated against practitioner feedback.
29 to 44 per cent of a therapist's formal job is already augmented.When mapped to O*NET occupational task structures, current AI usage covers roughly a third of a therapist's working week. In a sector facing persistent workforce shortages where additional headcount is not available, recovering a third of the administrative burden has substantial capacity implications across a team.
Information retrieval is the differentiator. 39 per cent of all AI usage involves retrieving and synthesising information from client records — not just writing. Summarising notes over time, looking up therapy records, checking client profiles, reviewing care plans. AI tools that only augment writing miss a substantial portion of where clinicians find value.
Usage deepens naturally with experience. Therapists begin with quick documentation tasks and expand to complex clinical reporting over weeks. The highest per-instance time savings (progress reports, NDIS documentation, care plans) appear later in the adoption cycle. Early deployment metrics understate the long-term value.
How we measured this
All data comes from Minikai's AI assistant (Mini), deployed across two NDIS disability care providers in Australia. The dataset includes 8,017 task instances. Data is current to 15 March 2026.
Every conversation with Mini is processed by an LLM-based evaluation pipeline that classifies it into one of 50 task types across four categories: writing assistance, information retrieval, decision support, and other. The pipeline also rates whether the AI successfully helped.
Time savings are LLM-estimated, not directly measured. They do not account for time spent reviewing or validating AI output, nor do they capture cases where output is discarded. Reported savings should be treated as indicative of magnitude rather than precise measurements.
To estimate what share of a therapist's working week AI covers, observed usage is mapped onto O*NET occupational task structures. O*NET data reflects U.S. occupational structures. Australian NDIS roles carry heavier documentation obligations not captured in O*NET, meaning actual AI usage is likely higher than these estimates. The sample is limited to two NDIS providers.
- Jobs and Skills Australia (2025). Vacancy Report, April 2025. jobsandskills.gov.au
- Kooijman, C.M. et al. (2025). Impact of using an AI scribe on clinical documentation and clinician-patient interactions in allied health private practice. Disability and Rehabilitation: Assistive Technology. doi.org/10.1016/j.dhjo.2025.101815
- Massenkoff, M. & McCrory, P.B. (2026). Nowcasting the Labor Market Impacts of Generative AI: Evidence from Observed Workplace Adoption. anthropic.com/research/labor-market-impacts
- National Center for O*NET Development (2025). O*NET 30.2 Database — Task Statements, Task Ratings. onetcenter.org