The admin load on Australian health, care, legal and claims operators: the numbers in 2026

Every operator we work with says some version of this: the work grows faster than the team. We wanted to know whether the public numbers agree. They point the same way, although no public series measures admin workload directly, and as far as we can tell nobody has put them side by side.

This report brings together the latest figures we could verify on 28 September 2026 from the NDIA, AIHW, ABS, Jobs and Skills Australia, the Productivity Commission, the ANAO, APRA, the RACGP and more than 30 other sources. Every figure carries a numbered citation to the source list at the end, the key passages are shown as screenshots of the original documents, and the data behind every chart and table can be downloaded as CSV. Where we have combined numbers ourselves, we label it Pragmatic AI analysis and show the method. Where a figure is thin or contested, we say so.

Key findings

• ▪The NDIS had 782,013 active participants at 30 June 2026, 67.6% more than five years earlier¹²⁴. It paid out $51.5 billion in 2025-26 on a cash basis¹, and paid 142.7 million claims in calendar 2024, about 390,000 a day³².
• ▪Between February 2021 and February 2026, employment in health care and social assistance grew 34.1%, while clerical and administrative employment across the economy grew 8.3% (Pragmatic AI analysis of ABS Labour Force data)⁶⁰⁶¹. These series do not measure admin workload directly.
• ▪Jobs and Skills Australia projects health care and social assistance jobs to grow 22.9% by 2035, but clerical and administrative jobs only 9.5%, with secretaries shrinking 5.7%¹⁴.
• ▪Only 2 of 71 clerical and administrative occupations are in national shortage, while GPs, registered nurses, disability support workers and solicitors are¹⁸¹⁹. Our reading is that the admin overflow tends to land on the people who are hardest to hire.
• ▪77% of GPs are dissatisfied with the amount of administration in their work, up from 58% in 2022²².
• ▪44% of WA doctors in training reported spending three or more hours of an eight-hour shift on admin (Pragmatic AI analysis of the survey's own bands)²³.
• ▪New NDIS appeals to the Administrative Review Tribunal rose 135% in two years, to 9,521 in 2025-26, while participants grew 18% (Pragmatic AI analysis)²⁵⁵¹.
• ▪Targeted NDIA pre-payment reviews covered about 0.4% of outlay value and cancelled more than half of the value they reviewed³². For Medicare, $44.7 million in MBS debt was raised (not necessarily recovered) across 3,779 cases in 2024-25³⁵; a 2023 review estimated MBS provider non-compliance at $1.5 billion to $3 billion a year³⁴.
• ▪12% of Australian businesses used AI in 2024-25, and 17% in health care and social assistance³⁸.
• ▪Published estimates of what AI could take on range from JSA's modelling assumption of 13% of tasks to McKinsey's 57% of US work hours in technical potential⁴⁷⁵⁰. None measures the routine share of Australian admin. On simple arithmetic, removing 13% of a team's human work gives it about 1.15 times its capacity, and removing half gives twice (Pragmatic AI analysis).

Case volumes grew by double digits while the admin workforce barely moved

The NDIS is the clearest case. The NDIA's June 2026 quarterly report to disability ministers counts 782,013 active participants with approved plans¹², up from 466,619 at 30 June 2021⁴. Total payments reached $51.5 billion in 2025-26 on a cash basis¹, up 12.2% on the year before⁵⁶ and 86% on 2021-22⁵³: about $141 million a day. The claim count is closer to the admin load: the NDIA paid 142.7 million claims in calendar 2024, about 390,000 a day³², and 68.8% of participants have a plan manager handling their invoices³.

Primary source · National Disability Insurance AgencyView original ↗

Health volumes grow more slowly, from a much larger base. The AIHW counts 12.8 million hospitalisations in 2024-25, up from about 11.8 million in 2020-21. 64% were same-day, rising to 75% in private hospitals⁵. The Department of Health, Disability and Ageing reports 482.5 million Medicare services in 2025-26 and $35.0 billion in benefits, or about 1.3 million services for every day of the year⁶. APRA's June 2026 figures show private health insurers paid for 1.35 million hospital episodes in the quarter, 3.1% more than a year earlier, with hospital benefits up 7.0%⁷⁵⁹.

Claims and legal work is growing unevenly. Safe Work Australia counts 146,700 serious workers' compensation claims in 2023-24, 34.5% more than a decade earlier⁸. State and territory civil lodgements have been flat at about 405,000 a year, but non-family-law lodgements in Division 2 of the Federal Circuit and Family Court, which include migration and other general federal matters, nearly doubled in two years, to 11,369 in 2024-25⁹. The Administrative Review Tribunal received 85,845 lodgements in 2024-25, 71% more than the year before¹¹.

Aged care is the newest pressure point. At 31 March 2026, 364,723 people were in the Support at Home ongoing services group, 338,049 of them receiving care and the rest assigned a place and still deciding, and another 100,191 were waiting in the priority system¹².

ABS Labour Force data shows employment in health care and social assistance grew 34.1% between February 2021 and February 2026, to 2.39 million people. By our calculation the industry supplied about a third of all jobs growth in that time. Clerical and administrative employment across the whole economy grew 8.3%. Measured from February 2019, before the COVID dip, it grew only 3.3%⁶⁰⁶¹. These are the February 2026 figures from the final Labour Force, Australia, Detailed release¹³, used as a fixed snapshot; later ABS revisions may differ slightly.

This is not a like-for-like comparison. Clerical workers serve every industry, not every new participant creates the same paperwork, and none of these series measures admin workload or who does it. But the direction is hard to miss: care volumes rose by double digits while the workforce that traditionally processes the paperwork barely grew.

The next decade widens the gap

Jobs and Skills Australia's projections to May 2035 expect health care and social assistance to add 541,900 jobs, a 22.9% increase and more than a quarter of all projected growth. Clerical and administrative workers are projected to grow 9.5%, below the 13.3% for all jobs. Receptionists grow 8.3% and accounting clerks 5.9%, while secretaries, the group that includes legal and medical secretaries, shrink by 5.7%¹⁴.

The hiring market is already moving that way. JSA's August 2026 Vacancy Report shows online job ads for clerical and administrative workers down 2.5% on a year earlier, the only major occupation group to fall¹⁵. Over the same period, ads for registered nurses rose 16.1%⁵⁷.

Labour is also getting dearer. The ABS Wage Price Index rose 3.2% across all industries in the year to June 2026 and 3.8% in health care and social assistance, the highest of any industry¹⁶. The Fair Work Commission raised modern award minimum rates by 4.75% from the first full pay period starting on or after 1 July 2026, with additional adjustments for the lowest-paid classifications¹⁷.

Pragmatic AI analysis, on an assumption we cannot test from public data: if the sector's admin need tracks its employment, it grows about 23% by 2035 while the economy-wide admin workforce grows about 9.5%¹⁴. The difference will be absorbed one of two ways. Either qualified staff do more of the admin themselves, or more of the routine work stops needing a person.

The shortage is in the people the admin lands on

JSA's 2025 Occupation Shortage List rates 29% of assessed occupations in shortage (293 of 1,022)¹⁸. Among 71 clerical and administrative occupations, only two were in national shortage: clinical coders and conveyancers¹⁹. Medical receptionists, health practice managers, legal secretaries, insurance loss adjusters and accounts clerks were all rated not in shortage¹⁹, and JSA notes that no clerical occupation was in persistent shortage from 2021 to 2025¹⁸.

Primary source · Jobs and Skills AustraliaView original ↗

The occupations that are short are the qualified ones. GPs, registered nurses in aged, acute and primary care, disability support workers and solicitors are all on the list, solicitors in every state and territory¹⁹. The Department of Health's nursing study projects an undersupply of 70,707 full-time-equivalent nurses by 2035, about 79,500 people²⁰. Its GP modelling projects a gap of 7,700 GPs by 2033²¹.

This is the pattern we think matters most, and it is our inference, not something these sources measure. The shortage list says which occupations are short, not who does the paperwork. In our experience, operators are rarely stuck because they cannot hire an admin officer. The admin their admin team cannot absorb tends to end up with the nurse, the GP, the solicitor or the support coordinator, who are the hardest people in the country to hire. Every hour of routine case admin on their desk is an hour of scarce capacity gone, and each operator can measure how many hours that is.

Nobody measures clinicians' admin time nationally, but several surveys report the pressure

There is no current national measure of how much of an Australian clinician's week goes to administration. The most recent broad estimate, from the MABEL survey of doctors, put non-clinical work at just under 7 hours a week, or 16% of working time²⁴. That was 2012 data, and it includes education as well as admin. What we have instead is a set of narrower surveys and studies, using different measures, that report administrative pressure.

The RACGP's Health of the Nation 2025 surveyed 2,416 GPs. 77% were dissatisfied with the amount of administration in their work, up from 58% in 2022, 60% in 2023 and 70% in 2024²².

Primary source · Royal Australian College of General PractitionersView original ↗

79% were dissatisfied with the time they spend on work Medicare does not fund. Asked which admin tasks they would change or remove, 44% named authority prescriptions²².

In hospitals, the AMA (WA) Hospital Health Check 2026 surveyed more than 1,300 doctors in training. Adding up the report's own bands, 44% reported spending three hours or more of an average 8-hour shift on administrative tasks²³. The report's text gives 39% for 2 to 4 hours, but its chart bands for that range sum to 41%, so we use the bands.

In nursing and aged care, a time-and-motion study by Westbrook and colleagues (University of New South Wales and University of Technology Sydney) observed 57 nurses on two wards of one Sydney teaching hospital and found they spent 37.0% of their time with patients in the first observation period and 35.7% in the second²⁵. The ANMF's August 2026 aged care survey found 46.6% of aged care workers rated the time they spend on paperwork as poor or very poor, rising to 50.9% of registered nurses²⁶. That is a rating, not a count of hours. An observational study of community aged care case managers found documentation took a third of their office time, rising from 29.4% to 37.0% after the consumer-directed care reforms²⁷.

Disability providers report the cost rather than the hours. National Disability Services' State of the Disability Sector Report 2025 surveyed 290 organisations: 48% ran a loss in 2024-25 and 81% of those delivering NDIS supports said they cannot keep doing so at current prices. Of 199 respondents to the question, 77% delivered unfunded services; the 149 that estimated the cost averaged more than $460,000 each. The report names NDIS administrative and regulatory issues among the causes²⁸.

We found no Australian survey of lawyers' or claims staff's admin time. The nearest benchmark is Clio's 2025 US data: lawyers recorded 3.0 billable hours in an 8-hour day, or 38%²⁹. The rest of the day is not all administration, and this is not an Australian sample. Overseas clinical studies are starker: a 2016 US study found doctors spent 27% of their office day with patients and 49.2% on records and desk work (Sinsky and colleagues)³⁰, and a 2025 UK study of resident doctors found 43.4% of their time went on admin against 11.7% on direct patient contact³¹.

The backlog is where the admin shows up

The NDIA's own service guarantee figures show where the queue sits. Access has recovered: 96% of access decisions or requests for more information were made within 21 days in the June 2026 quarter, against 9% in the March 2025 quarter. But only 32% of participants who asked for a plan reassessment got a decision on whether it would go ahead within 21 days (86% of accepted reassessments were then completed within 28 days), 44% of plan amendments were made within 28 days and 48% of internal reviews were finished within 60 days².

Appeals follow. By the NDIA's count, new NDIS cases at the Administrative Review Tribunal went from 4,044 in 2023-24 to 7,137 in 2024-25 and 9,521 in 2025-26².

The Tribunal's 2024-25 annual report says 71% of NDIS matters it finalised were resolved by consent, meaning the parties agreed an outcome without a hearing. The Tribunal and the NDIA count NDIS applications differently, so we use the NDIA's count above. Across its whole caseload, in which migration and protection matters are the largest part, the Tribunal says plainly that it does not have "the staff, member and technological resources" to manage it¹¹.

Primary source · Administrative Review TribunalView original ↗

The pattern repeats elsewhere. In Division 2 of the Federal Circuit and Family Court, 11,598 of 20,508 pending non-family-law matters (56.6%) were more than 12 months old at 30 June 2025⁶³. Elapsed time is not necessarily court delay, but it is time a matter sits open. The median wait from approval to being assigned a home care package was 204 days in 2024-25, up from 78 days a year earlier¹⁰.

Not every backlog is admin. Tribunal queues depend on member numbers and elective surgery waits (a median of 45 days in 2024-25, with 6.0% of patients waiting more than a year)⁵ depend on theatres and surgeons. But many of the decisions in the NDIS queues involve reviewing documents: reading evidence, checking it against rules and writing it up.

Few claims are checked, and targeted reviews find substantial problems

The ANAO's June 2025 audit of NDIS claim compliance reports an NDIA estimate that 6% to 10% of outlays could be non-compliant, fraudulent or incorrect: $2.0 billion to $3.5 billion in 2022-23. In 2024 the NDIA paid 142.7 million claims worth $42.9 billion, and reviewed 68,235 of them before payment. Those targeted manual reviews covered about 0.4% of outlay value in the latest quarters (0.3% across 2024), and more than half of the value they reviewed was cancelled. In a separate targeted analysis of $35 million of claims over $5,000, 41.2% of the claim value was rejected³². Because the reviews target risky claims, these are not failure rates for the scheme as a whole.

Primary source · Australian National Audit OfficeView original ↗

In May 2026, the NDIA told the parliamentary inquiry into NDIS integrity that it estimates broad "integrity leakage" at 8.2% to 8.3% of scheme payments, around $3.7 billion on about $45 billion a year⁶²³³. That covers administrative errors and payment mistakes as well as deliberate non-compliance and fraud; it is not a fraud estimate.

Medicare has the same shape. The 2023 Independent Review of Medicare Integrity and Compliance estimated provider non-compliance at $1.5 billion to $3 billion a year³⁴, and the ANAO repeated that range in a September 2026 audit. The same audit reports $44.7 million in MBS debt raised in 2024-25 against 3,779 identified cases of fraud and non-compliance³⁵. Debt raised is not money recovered, and the two figures have different bases, so they cannot be read as a recovery rate. But the gap in scale is plain. The widely reported $8 billion figure is contested: the review found it rests on a much broader definition, drawn from a small and unrepresentative billing sample, and only about $3 billion of it matches the department's definition³⁴.

Primary source · Australian National Audit OfficeView original ↗

Errors run the other way too. By 30 June 2025, Bupa had paid $14.3 million on more than 4,100 claims where members were wrongly told they could claim nothing when part of their treatment was covered³⁶, and on 11 December 2025 the Federal Court ordered it to pay $35 million in penalties⁵⁸. Services Australia reports that 1.9% of its 466 million claim outcomes in 2024-25 contained a critical error, and about 6% of social security claims did³⁷.

For an operator, the lesson is not about fraud. Many of the problems these audits describe are ordinary: missing evidence, the wrong item, a claim outside the plan, a late lodgement. Some of those checks can be written down as rules; disputed evidence and exceptions need human judgement. People check only a sample because checking everything by hand is unaffordable. Software can apply the written rules to every case and send the doubtful ones to a person.

AI adoption is broad in surveys and shallow in practice

The ABS Characteristics of Australian Business survey found 12% of businesses used AI in 2024-25, up from 1% in 2021-22. The figure was 35% for large businesses and 17% in health care and social assistance. The ABS notes the question does not measure how intensively AI is used³⁸.

The National AI Centre's adoption tracker, which asks SMEs to describe their stage of adoption, puts the figure at 43% for December 2025 to February 2026, with health among the sectors above half. Among businesses not adopting, about 65% cited distrust of AI decision-making or a preference to keep human control³⁹. That concern is reasonable, and it is a design requirement, not an obstacle.

In clinical work, The Medical Republic reported that 18.7% of 1,535 GPs in a May 2026 Healthed survey said they personally use an AI scribe⁴⁰; we could not check the survey's method. A self-selected RACGP newsGP poll put use in respondents' practices at 40%⁴¹, which measures something different. Among respondents to the 2025 Victorian Lawyer Census, 36.7% reported using AI tools, and 64.8% of those relied on general-purpose chat tools⁴². The best trial evidence on scribes is modest: in a 238-doctor randomised trial published in NEJM AI in 2025, one scribe cut time spent on notes by 9.5% and the other made no significant difference⁴³.

The rules are clear enough to work within. The OAIC's guidance says privacy obligations apply to what goes into an AI system and what comes out, and recommends against entering personal information into publicly available generative AI tools⁴⁴. Ahpra says practitioners remain responsible for care and must apply human judgement to AI output⁴⁵. The December 2025 National AI Plan relies on existing, technology-neutral law and regulators, rather than proceeding with the 2024 proposal for mandatory guardrails⁴⁶.

The adoption surveys do not measure how many complete cases AI handles. Our view from practice is that most use is still individuals using general tools on single tasks. That helps at the margin but rarely changes how many cases a team can finish, because a person still has to check and complete each case.

Estimates of what AI could take on vary widely, and none measures your admin

JSA finds automation potential is concentrated in routine clerical and administrative roles⁴⁷. The Productivity Commission expects multifactor productivity gains above 2.3% over a decade, worth about $116 billion of GDP over those ten years⁴⁸. McKinsey names legal and administrative services among the most exposed roles and is explicit that its figure is technical potential, not a forecast⁵⁰.

The honest reading is that the two Australian government models assume AI takes on a modest share of tasks across the whole economy, and that routine clerical work sits near the top of the exposure rankings. Neither tells an operator what share of their own cases could be finished without a person. That share can be far above or below any national figure, and it has to be measured on real cases, not assumed from a national study.

For a typical operator, the gain depends on how much human work actually goes

A team's capacity follows the share of its human work that is removed. On paper it does not matter how that happens: removing 13% of the work in every case frees the same time as finishing 13% of equal-effort cases with no one touching them.

Our method: capacity = 1 ÷ (1 − the share of human work removed). It assumes the remaining work takes the same effort and nothing else becomes the bottleneck. Remove 13% of the work, the figure JSA assumes for tasks in its economy-wide model⁴⁷, and a team has about 1.15 times its capacity. Remove half and it doubles.

An illustrative example: a team at full stretch handling 1,000 cases a week. If 13% of its work goes, for example because 130 of those cases finish without anyone touching them, the same team can take on about 1,150 a week. If half its work goes, it can take on about 2,000. These are round numbers to show the arithmetic, not a forecast.

In practice, and this is our view from our own work rather than a measured result, step-level help often removes less work than it appears to. A person still opens, reads, checks, hands off and closes every case, so saving a few minutes inside each one leaves most of the handling in place. A case that finishes untouched removes all of it. That is why we count whole cases.

We also think, and again it is a view rather than a statistic, that well-defined case types in these sectors carry more rule-based work than economy-wide models assume. The admin GPs most want removed is authority prescriptions and government and insurance forms²². NDIS claims are cancelled for not meeting written claim requirements³². Medicare non-compliance is, by definition, billing that breaks the rules³⁴. Invoices checked against a plan are exactly this kind of work. That is why we measure the untouched share on each operator's own cases before promising anything, and why the cases that need judgement stay with their staff.

How we built this report

• ▪Every figure comes from a public source available on 28 September 2026. We identify the release and period used for each series, and each figure carries a numbered citation to the source list below.
• ▪Figures labelled Pragmatic AI analysis are simple arithmetic on published numbers, and the method is stated beside each one and in the notes column of the CSV files.
• ▪Series are not always aligned: NDIS counts are at 30 June, ABS employment at February. Health care employment is seasonally adjusted, and clerical employment is original data because that is how the ABS publishes it. The ABS ceased Labour Force, Australia, Detailed after its March 2026 release, so we use its February 2026 workbooks as a fixed snapshot; later revisions may differ slightly.
• ▪The screenshots are crops of the original documents with the quoted passage highlighted. Each links to the document it came from.
• ▪We left out figures we could not trace to a primary source, including the claim that GPs spend up to nine hours a week on red tape (a 2013 AMA item with no published method) and the claim that lawyers spend 48% of their time on admin.

Each row gives the metric, value, unit, period, geography, source title, source URL and notes. Derived rows are marked DERIVED with their method. The data is free to reuse under CC BY 4.0 with a link to this page. We plan to update these numbers each year.

If you run one of these operations and want to know what share of your own cases is routine, start with a free conversation about how your cases arrive today.