Real-world data on how therapists use AI in their work is rare. Most public reporting comes from adoption surveys. This article is a step toward filling that gap.
Drawing on 6,224 AI interactions by 143 therapists using Minikai across multiple NDIS providers, this article looks at what they use AI for, how it informs their clinical work, where it saves them time, and what share of their workload AI assists with.
This article examines 6,224 real AI interactions by 143 therapists using Minikai across multiple NDIS providers. The data is current as of 1 May 2026. Rather than modelling what AI could theoretically do, we show what therapists do when using Minikai.
Minikai is an AI-native clinical support system built for the disability and aged care sector. Minikai maintains rich context for each participant, their history, preferences, goals, and comprehensive records of the care they receive, helping therapists work with them more effectively. We design and test the platform in partnership with providers, with a focus on quality of care.
Minikai is used by disability and aged care providers across Australia and supports a wider set of roles and use cases than the ones discussed here, including support workers, quality and governance teams, and others. The findings in this article are drawn from therapist use of Minikai across multiple NDIS-funded disability providers.
The workforce problem
Under the NDIS, therapists are funded to support participants' quality of life. This includes occupational therapists, physiotherapists, speech pathologists, behaviour support practitioners, nutritionists, exercise physiologists, and social workers. These therapists sit within the broader allied health workforce, which is facing a national shortage in Australia, including occupational therapists, and to a lesser degree speech pathologists and physiotherapists (Foster et al., 2025).
The nature of NDIS work adds to this. Supports delivered must be planned, evidenced, and reported against a participant's funded goals. That documentation load comes out of time that could otherwise be spent with participants.
AI will not solve the workforce shortage by replacing therapists. We are seeing that AI absorbs the documentation and administrative burden, returning time to care and concentrating therapists' time on the work that needs their expertise: clinical judgement, evidence-informed practice, and direct interaction with participants. The question is, how are therapists using AI?
What therapists use AI for
We have used AI to identify the tasks therapists perform when using Minikai, and for each interaction estimate how much time Minikai saved them on manual work. This estimation is grounded in user testing where therapists report on time-saved during their interactions. The distribution is steep, a small number of tasks account for most of the time saved, with a long tail of tasks performed less often. The top 20 tasks ranked by total hours recovered:
| Task | Category | Count | Hours saved | Median time |
|---|---|---|---|---|
| Edit or reformat text | Writing | 3,628 | 579 hrs | 5 min |
| Write progress reports and reviews | Writing | 594 | 389 hrs | 45 min |
| Draft emails to families/providers | Writing | 1,011 | 303 hrs | 12 min |
| Summarise notes over time | Info retrieval | 786 | 266 hrs | 15 min |
| Look up therapy and medical records | Info retrieval | 816 | 206 hrs | 15 min |
| Create notes (e.g. SOAP) | Writing | 572 | 196 hrs | 20 min |
| Write NDIS documentation | Writing | 261 | 164 hrs | 35 min |
| Summarise therapy or family contact | Info retrieval | 500 | 153 hrs | 15 min |
| Recommend care interventions | Decision support | 413 | 151 hrs | 20 min |
| Write assessments | Writing | 239 | 149 hrs | 35 min |
| Draft clinical handover documents | Writing | 157 | 104 hrs | 45 min |
| Look up assessments | Info retrieval | 537 | 103 hrs | 10 min |
| Write equipment funding justification | Writing | 151 | 98 hrs | 35 min |
| General overview of person | Info retrieval | 260 | 91 hrs | 20 min |
| Look up client profile and demographics | Info retrieval | 784 | 82 hrs | 3 min |
| Look up equipment and assistive tech | Info retrieval | 402 | 45 hrs | 5 min |
| Look up care requirements | Info retrieval | 266 | 43 hrs | 10 min |
| Write care plans | Writing | 64 | 36 hrs | 30 min |
| Write behaviour support plans | Writing | 53 | 35 hrs | 30 min |
| Summarise incidents or behaviours | Info retrieval | 77 | 26 hrs | 15 min |
Workflows typically start with therapists analysing information about a participant from across notes, assessments, plans, and prior reports, summarising what has happened over time, drawing connections across different information sources. That context is then used to write better-informed reports, emails, plans, and notes. Retrieval comes first, writing follows.
Within a single interaction, Minikai typically saves around 45 minutes when drafting a progress report, 45 minutes on a clinical handover, and 30 minutes on a care plan. These are the tasks where the most manual effort is taken out of a single sitting. High-frequency information retrieval tasks save less each time but the totals add up.
Time saved is not the time it took for the AI to do the work. It is time the therapist did not spend pulling records, drafting structure, summarising history, or reformatting output. The clinical reasoning stays with the therapist.
The figures do not capture cases where the AI output is discarded entirely or where Minikai was used for tasks that would not have been done manually otherwise.
Total hours recovered across the top 20 task types over 6,224 AI-assisted interactions.
Time saved per therapist
Time saved is not evenly spread across the therapist cohort. The breakdown for March 2026 is below.
More active users save more time. About half the cohort saves under 5 hours per month, while a smaller group of intensive users saves 20 hours or more. The 5 to 20 hour band, where most engaged therapists sit, accounts for the majority of total time saved. The lower bands largely reflect therapists still embedding Minikai in their daily workflows, and we expect more to move into the higher bands as adoption deepens.
What this looks like in practice
The examples below show what AI-assisted work looks like across a working session.
Functional Capacity Assessment, Occupational Therapist
An occupational therapist (OT) is preparing a Functional Capacity Assessment (FCA) for a participant with complex needs. NDIS funding for an FCA might cover eight hours, but in practice these can take 16 hours of work, time the provider absorbs unbilled. The OT uses Minikai to summarise the participant's goal history, prior assessments, and recent session notes, then draft structured sections against the FCA template, with citations back to the source records. The OT does the clinical reasoning, adjusts language, and finalises the report in less time.
Equipment funding check, Occupational Therapist
An OT is recommending a new wheelchair for a participant. Before specifying the chair, the OT asks Minikai whether the participant has remaining funding in their assistive technology category, and what equipment recommendations have already been made in prior assessments. Minikai surfaces the budget position and prior recommendations from the record, allowing the OT to specify a chair within the participant's funding.
Justification for increased daily living support, Behaviour Support Practitioner
A behaviour support practitioner (BSP) is building the case for increased daily living support funding ahead of a participant's plan review. The BSP uses Minikai to summarise behavioural patterns across the last six months, surface incident frequency and severity, and pull together evidence of unmet need from session notes and incident reports. From that retrieved view, they ask Minikai to draft an evidence-based justification linking observed behaviours to the requested support increase. The BSP validates the clinical reasoning and finalises. The justification supports an increase at plan review.
Session preparation, Speech Pathologist
A speech pathologist is preparing for a dysphagia review. They ask Minikai for a quick brief on the participant: mealtime management plan, recent nursing notes on swallowing observations, and the last assessment results. The speech pathologist arrives at the session with full context instead of spending 30 minutes pulling records. After the review, the speech pathologist updates the participant's IDDSI food and drink levels and asks Minikai to draft new mealtime instructions for the support workers, citing back to the underlying assessment.
How much of a therapist's job does AI assist with?
Beyond individual tasks, the question is what proportion of a therapist's total work AI assists with. To answer this, observed Minikai task usage is mapped onto O*NET tasks across therapy roles. O*NET is a U.S. Department of Labor dataset that defines the core tasks of an occupation and how often each is performed. Therapy roles vary across countries, so the U.S. data approximates the Australian role rather than a direct match.
For each O*NET task in a role, we check whether Minikai assists with it, then weight by how often the task is performed. A task done daily counts more than one done yearly. The resulting figure is a share of tasks weighted by frequency.
Across the seven therapist disciplines, AI assists with between 23 and 55 per cent of their role. AI does not replace direct interactions with participants.
What this means for providers
Increased revenue. Providers face pressure to recover more billable time from each therapist. AI helps recover billable time on tasks that often run longer than funded, such as Functional Capacity Assessments, which therapists have anecdotally told us can take up to twice the funded hours. We have observed therapists using Minikai to close that gap, and providers have reported increased utilisation among their therapists, increasing revenue.
Better-informed clinical work.Information retrieval is one of the most frequent uses of Minikai and feeds the writing tasks that follow. Therapists are using Minikai to surface what is already documented in a participant's record before making decisions, applying for or justifying funding with evidence, and investigating incidents, rather than relying on recall or the most recent note. The clinical reasoning still belongs to the therapist.
Improved staff experience and lower risk of burnout.Qualitative research on physiotherapists and occupational therapists working under the NDIS identifies onerous NDIS reporting requirements, billable-hour expectations, and unpaid overtime as central contributors to burnout in the disability sector (Currie & Dafny, 2025). Some therapists using Minikai have anecdotally reported finishing their work earlier.
Capacity in a workforce-constrained sector. A national shortage of allied health professionals persists in Australia (Foster et al., 2025). In a sector where workforce supply is constrained, time recovered across the 23 to 55 per cent of tasks AI assists with is capacity providers can choose to redirect.
Notes
This article includes information from the O*NET 30.2 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA). Used under the CC BY 4.0 license. O*NET® is a trademark of USDOL/ETA. Minikai has modified some of this information. USDOL/ETA has not approved, endorsed, or tested these modifications.
- Currie, M. & Dafny, H.A. (2025). Burnout in physiotherapists and occupational therapists working in the Australian community disability sector under the National Disability Insurance Scheme: a qualitative study. Disability and Rehabilitation, 47(23), 6122-6134. doi.org/10.1080/09638288.2025.2476022
- Foster, A.M., Munro, S., Golder, J. & Mitchell, D. (2025). Why they come, why they stay and why they leave: a survey to understand the drivers of recruitment, retention, and attrition of allied health clinicians in an Australian metropolitan health network. BMC Health Services Research, 25, 767. doi.org/10.1186/s12913-025-12922-3
- National Center for O*NET Development (2026). O*NET 30.2 Database, Task Statements, Task Ratings. onetcenter.org/database.html