Empathia Group · SIL economics · The SIL house cost model
The real unit economics of SIL
The NDIS prices an hour of support. A SIL provider is paid in fractions of a worker and has to roster whole ones. This paper sets out the arithmetic of a SIL house, names each way a house loses margin and how to measure it, and explains why pooling participants' plans should not be used to close the gap.
11%
of the required roster is vacant or unfunded, on average, across the roster patterns we examined
8.4%
is the margin the price limits leave the same houses when every hour is funded and they are run as the cost model assumes
6 in 10
houses have every place filled and ratios that add up to whole workers; the unfunded hours sit in the other four
1 in 10
residents is supported at a single ratio for the whole week; the typical resident moves between three
From anonymised roster patterns in more than 100 rosters of care. The patterns show how support is shared and do not identify plan funding. Figures taken from them are rounded.
In brief
The price is built for one worker supporting one person. In a shared house the plans pay for fractions of a worker and the roster needs whole workers. The two match only when the ratios of the residents sharing a worker add up to that worker, hour by hour.
A rostered hour that no plan pays for is a vacancy, unfunded support or a plan overrun. The three have the same effect on a house and different causes, records and owners. Comparing what is funded, what the roster of care requires and what was rostered tells them apart, and separates all three from losses made in rostering.
In the roster patterns we examined, vacancy and unfunded support take about 11 percent of the required roster, valued at the price limit. A fully funded house with the same mix of sleepover and awake nights earns a margin of 8.4 percent when it is run as the cost model assumes, so across these houses the roster of care alone takes more than the price provides, before operations are considered.
The loss is concentrated. The median house has no vacant or unfunded hours, and one house in ten has more than a third of its hours vacant or unfunded.
Pooling the plans in a house removes the loss from the accounts by charging some participants for support delivered to others. We advise against it.
01What the price pays for
The NDIA sets SIL price limits with its Disability Support Worker Cost Model, which is a cost plus overheads model with a two percent margin. A standard weekday hour is priced at $73.58, of which $52.94 is the worker's wage with leave, superannuation and allowances. The worker costs about 72 percent of the price, and everything else is met from the remaining 28 percent.
Exhibit 1
What a $73.58 hour pays for
NDIA Disability Support Worker Cost Model 2026-27, DSW Level 1, weekday daytime.
52%
20%
16%
11%
■ Base wage 52.3%■ Leave, super and allowances 19.6%■ Operational overheads 15.6%■ Corporate overheads 10.5%■ Margin 2.0%
Layer
Per hour
Of price
What the model assumes
Base wage
$38.50
52.3%
SCHADS 2.3
Leave, superannuation and allowances
$14.44
19.6%
37.51% on the base wage
Operational overheads
$11.46
15.6%
21.65% on the loaded wage, published as one figure
Non-billable time
$5.15
7.0%
A worker bills 92% of paid hours: breaks 4.17%, training 3.29%, other 0.54%
Supervision
$3.94
5.4%
One supervisor at SCHADS 3.2 to fifteen full-time equivalent workers
Workers compensation
$1.67
2.3%
About 3.6% of pay including leave (1.7% in 2021-22)
Casual loading
$0.70
0.9%
70% of the workforce permanent
Corporate overheads
$7.73
10.5%
12% on all of the above
Margin
$1.44
2.0%
2% on all of the above
The four components of operational overheads are an Empathia reconstruction. The 2026-27 model does not itemise them. The 2021-22 model did, and its parameters reproduce the 2026-27 figure for DSW Levels 1 to 3 with workers compensation at 3.57%.
The model publishes operational overheads as one figure. The breakdown in Exhibit 1 is our reconstruction from the 2021-22 edition, whose parameters reproduce the current figure once workers compensation is raised from 1.7 percent to about 3.6 percent. It is an inference and is not published by the Agency. Four points about the model carry through the rest of this paper.
Its parameters describe an efficient provider. Each was set at the level achieved by the most efficient quarter of providers in the Agency's benchmarking survey. The survey averages were 79.8 percent utilisation and one supervisor to 11.8 workers, so a provider at the sector average begins below the model's margin.
Utilisation refers to a worker, not a house. It is the share of a worker's paid time that can be billed, set at 92 percent for the worker level used in SIL, with the rest allowed for breaks, training and other non-billable time. The model has no allowance for a roster that is only partly funded.
Supervision is priced per full-time equivalent. One supervisor to fifteen full-time equivalent workers is about one supervisor to twenty-five people at the sector's average hours.
Every layer scales with the penalty rate. The worker's share of the price is about 72 percent in every time band, so the amount left for everything else rises from $20.64 an hour on a weekday to $37.05 on a Sunday. Supervision provided on a Sunday does cost twice as much, because the model pays supervisors the same loadings as the workers they supervise. Corporate overheads do not cost more on a Sunday, and they rise from $9.17 to $16.64 an hour.
02Why a house does not behave like an hour
A SIL plan funds its participant for a fraction of a worker, and a house is rostered in whole workers. A resident funded at one to three is billed a third of a worker for each hour of support. Three such residents fund exactly one worker for as long as all three are at home and sharing that worker. If one of them is at a day program, the two at home fund a whole worker only if their plans pay one to two for those hours.
In every hour of the week, the whole workers rostered equal the fractions billed to residents, plus the fractions attached to vacant places, plus a remainder that no plan covers. The unfunded part has four sources.
A vacant place. A vacancy in a one to three house leaves a third of a worker unpaid, and the roster cannot be cut by a third of a worker.
Ratios that do not sum. A resident funded at one to one who shares a house with a resident funded at one to three needs two workers if each is rostered at their funded ratio, and their plans fund one and a third.
Changes in who is sharing. When one of three residents leaves for a day program, or receives one to one support to get up or to eat a meal, the other two are sharing a worker between two people. That is funded only if their plans step up to one to two for those hours.
Plans of different sizes. A house rostered at one to three needs each plan to pay for a third of the roster. A small plan may not cover its third, and a large plan cannot make up the difference, because each plan pays only for its own participant.
Exhibit 2
Eight houses at the same prices
The part of each house's roster that no plan pays for, the name it goes by, and the house's revenue and margin when it is run exactly as the cost model assumes. Select a house to see where each dollar it bills goes.
House
Unfunded a year
Shown as
Revenue a year
Margin
+Three residents at 1:3 Sleepover. Every place filled.
0
Fully funded
$626k
12.7%
+The same house with one place vacant One worker rostered, two thirds of a worker funded.
$209k
Vacancy
$417k
−30.9%
+A resident funded at 1:1 with a resident funded at 1:3 Each rostered at their funded ratio: two workers, one and a third funded.
$417k
Unfunded support
$835k
−30.9%
+The same two residents sharing one worker, both claimed at 1:3 One worker, two thirds funded. Most of the larger plan is unspent.
$209k
Unfunded support
$417k
−30.9%
+The same two residents sharing one worker, both claimed at 1:2 One worker, claimed in full. The smaller plan cannot pay for a half.
$104k
Plan overrun
$522k
−4.8%
+Three at 1:3 through a real week, plans step up Two out on weekdays, two with morning one to one support.
0
Fully funded
$720k
11.3%
+The same week, plans stay at 1:3 The same roster, with the changes in ratio unfunded.
$108k
Unfunded support
$612k
−4.3%
+Two residents at 1:1, awake overnight No sleepover, and no ratio to leave a remainder.
0
Fully funded
$1,566k
2.0%
Unfunded is valued at the price limit: the billing that the vacant place, the ratio remainder or the support above a plan would have produced. Every house bills at the same price limits.
The margins in Exhibit 2 are for houses run exactly as the cost model assumes, so the differences between them come from the houses alone.
The cost of a vacancy
One vacant place moves the first house from a margin of 12.7% to a loss of 30.9%. The house loses about $209,000 of billing a year and none of its cost. The roster, its supervision and the back office are unchanged, so the whole of the lost billing comes off the result.
Residents funded at different ratios
The third, fourth and fifth houses are the same two residents, one funded at one to one and one funded at one to three. The provider has three ways to roster and claim for them within the rules, and none is funded in full.
Roster each resident at their funded ratio. The house needs two workers and the plans fund one and a third. Two thirds of a worker is unfunded support, and the house loses about $258,000 a year on revenue of $835,000.
Share one worker and claim both residents at one to three. One worker is rostered and two thirds is funded, so the unfunded support falls to a third of a worker and the loss to about $129,000. Revenue halves, to $417,000, and two thirds of the larger plan is left unspent, which puts that funding at risk at the next plan review.
Share one worker and claim both residents at one to two, the ratio at which the support is delivered. The roster is whole and shows no unfunded support. The resident funded at one to three is receiving more than their plan pays for, and the difference, about $104,000 a year, is a plan overrun: the smaller plan runs out before the plan period ends and the provider carries the rest. The house loses about $25,000.
These options are stated as hourly claims, where each plan is charged for the support its participant received. Section 03 sets out how the same arithmetic appears under weekly claiming.
The two plans together hold $835,000 a year, and one shared worker is worth $626,000 at the price limit. The house is funded in total and still loses money under every compliant arrangement, because the funding sits in the wrong plan.
The third option is unfunded support under another name. Take a single participant funded at one to two who needs, and receives, one to one support. The provider is paying for half a worker that the plan does not fund. Claimed at one to two, the half appears on the roster as unfunded support. Claimed at one to one, the roster is whole and the same half appears as a plan overrun, worth about $313,000 a year in a sleepover house either way. The loss is the same in both cases and only the record it appears in differs, so a provider needs to know which record to look in.
The sleepover, and how much a house can carry
The first house earns 12.7% on a sleepover and 2.0% when it is staffed awake overnight, and 2.0% is the cost model's own margin. A sleepover is billed at $311.79 a night against an award allowance of $62.87, which makes it the largest source of surplus in SIL. Providers know this, and in practice the sleepover pays for vacancy, unfunded support and rostering losses elsewhere in the portfolio.
A house can carry unfunded hours up to its fully funded margin and no further. For a house staffed awake overnight that is two percent of its roster, or about 3.4 hours a week for each worker on shift. Awake houses are usually at one to one or two to one, where a vacancy releases a whole worker and the ratios rarely leave a remainder, so they earn less and vary less. For the one to three sleepover house the limit is 12.7%. Its sleepover surplus is worth about $67,000 a year, and one vacant place costs $209,000, so a single vacancy uses the sleepover surplus of about three fully funded houses.
03Plans are patterns, and they differ in size
The third and fourth sources account for most of the complexity in a roster of care and are the least widely understood.
A plan funds a pattern of support
A plan does not fund a single ratio. The ratio changes whenever the group sharing a worker changes, which happens when a resident is out of the house and when a resident receives one to one support. In the roster patterns we examined:
9 percent of residents have one ratio for the whole week, and the median resident has three;
82 percent have day program or employment hours in their roster;
in the median shared house, some residents are being supported while others are out for 14 percent of the week;
one resident in five has an average ratio that is close to a standard ratio.
Exhibit 3 follows three residents at one to three through a week in which two attend a day program and two receive one to one support in the mornings. If their plans step up when the sharing group shrinks, every rostered hour is funded and the house earns 9.8% at the settings of a typical provider. If the plans stay at one to three, 20% of its rostered hours are unfunded support and it loses 6.1%. The residents and the roster are the same in both cases.
Exhibit 3
A house through a week
Opens on three residents at 1:3, two of whom are out for six hours each weekday and two of whom have one to one support in the mornings. Change the house to see where its rostered hours go.
A weekday is 6am to 8pm, a weekday evening is 8pm to 10pm, and the night is 10pm to 6am. An awake house bills weekend nights at the weekend rate. The sharing group shrinks when a resident is out or is receiving one to one support. Vacant hours and remainder hours are both rostered hours that no plan funds.
Plans in the same house are different sizes
Take three residents with plans of $100,000, $300,000 and $500,000 in a house rostered at one to three. Each resident's third of the roster costs about $209,000 a year. The smallest plan cannot pay for its third, the largest has $291,000 left over, and the surplus in one plan cannot be applied to the shortfall in another. The shortfall is a plan overrun: the resident receives a third of a worker and the plan runs out before the year does.
Exhibit 4
One house, three plans
Three residents rostered at 1:3 on a sleepover, with plans of $100,000, $300,000 and $500,000, at the settings of a typical provider.
Resident
Plan
Their third of the roster
Short
Unspent
Resident 1
$100k
$209k
$109k
0
Resident 2
$300k
$209k
0
$91k
Resident 3
$500k
$209k
0
$291k
Approach
Plan overrun a year
Plan funding unspent
Margin
Each plan pays for its own third of the roster
$109k
$383k
−7.5%
Resident 3 also has four hours a day of one to one support, and the others step up for those hours
$130k
$278k
−9.4%
The loss here is a plan overrun: the part of a resident's share of the roster that their plan cannot pay for. The funding left over in the other two plans cannot be applied to it.
The provider has few options within the rules. Raising the ratio for the resident with the largest plan uses more of that plan, and it also requires the other two residents to step up for those hours, which deepens the overrun in the smallest plan. The second row of Exhibit 4 shows this.
Claimed by the hour, the house also sends two different signals. The Agency sees $383,000 of unspent funding, in plans that together exceed the value of the roster. The provider sees a loss of 7.5%. Both readings are accurate, and neither party can see the other's from its own records.
The same house under weekly claiming
In our experience about nine SIL providers in ten claim weekly. A weekly claim is the participant's regular SIL funding divided by the weeks in the plan, made against a service agreement that records the agreed typical schedule of supports. The comparisons above still apply, and they show up differently.
A plan overrun appears as a schedule that costs more than the weekly amount. The provider delivers the agreed schedule and is paid the weekly amount, so the smallest plan in Exhibit 4 still leaves it $109,000 a year short.
Unspent funding does not appear. The weekly amount is the funding, so the two larger plans would be claimed in full, $383,000 a year above the value of the third of a worker each of their participants receives. That claim is supported only if the schedules behind those plans are delivered, which means rostering the additional support the plans fund. A provider that shares one worker between the three residents and claims the three weekly amounts is pooling.
The Agency loses its signal. Under weekly claiming every plan is fully used by construction, so the Agency sees neither unspent funding nor a shortfall in this house. What separates a compliant provider from a pooled one is whether the schedule behind each weekly amount was delivered, and that can be seen only by reconciling the schedule to the roster.
The constraint eases as a house grows, because more residents allow more groupings at different ratios, and it does not arise in one to one houses. Resolving it takes careful matching of residents and a roster of care built around each plan. It can be done: we have seen providers resolve buildings of ten residents with widely varying plans through close attention to the roster of care.
04Measuring the loss: funded, required, rostered
A vacancy, unfunded support and a plan overrun have the same effect on a house: a worker is paid, and no plan pays for part of that worker. They are worth separating because each has its own cause and its own remedy. A vacancy is filled by a placement, unfunded support is resolved in the roster of care, and a plan overrun is resolved in the plan. Rostering losses are different again, and are fixed by the people who build the roster each week.
Telling them apart requires fixed points to measure against. There are three, and they can be worked out for any house from its roster of care and its payroll.
Funded. The sum of the fractions of a worker that the residents' plans pay for.
Required. The whole workers the roster of care needs once those fractions are rounded.
Rostered. The workers who were paid, taken from payroll.
Exhibit 5
Funded, required, rostered
One half hour in a house with two residents funded at 1:3 and one at 1:2.
Level
What it measures
In this half hour
Workers
Funded
The fraction of a worker the plans pay for
1/3 + 1/3 + 1/2
1.17
Required
The whole workers the roster of care needs
One for the two at 1:3, one for the 1:2
2.00
Rostered
The workers who were paid, from payroll
For example
2.10
Gap
Workers
What it is
Owner
Required less funded
0.83
Created by the roster of care, before anyone is rostered. Where a place is empty it is a vacancy obligation, and otherwise it is unfunded support.
Occupancy, or the plans and the roster of care
Rostered less required
0.10
Created in rostering. It can be negative, when fewer hours are worked than the roster of care requires.
Operations
Each gap has one cause and one owner. Measured for every half hour of a week, the two gaps separate a poorly constructed roster of care from a poorly run roster.
Required less funded is the loss built into the roster of care before anyone is rostered. It is a vacancy obligation where a place is empty and unfunded support otherwise, and the two are separated by giving each vacant place its own service pattern. Rostered less required is a rostering result. Because it is measured against the required roster and not against funding, a rostering problem and a funding problem cannot be mistaken for each other. A further comparison, between the roster of care priced over the plan period and each plan's budget, identifies plan overrun and unspent funding.
This is the logic of Empathia's SIL Dashboard, which makes the comparison for every half hour of a roster of care. The method does not depend on the tool, and the definitions in Table 1 at the end of this paper can be applied to any set of rosters.
Exhibit 6
What the roster patterns show
Required staff hours in anonymised roster patterns from more than 100 rosters of care.
■ Covered by the residents' ratios about 89%■ Vacancy obligation about 5%■ Unfunded support about 6%
Houses
Every place filled and no unfunded support
about 60%
Vacant and unfunded share of required hours, median house
none
The same share, 75th percentile house
about 20%
The same share, 90th percentile house
about 35%
Staffed awake overnight
about 35%
Residents
One ratio for the whole week
about 10%
Ratios in the median resident's week
3
Day program or employment hours in the roster
about 80%
Average ratio within 3% of a standard ratio
about 20%
Source: Empathia RoC Optimiser, November 2026. The data are anonymised roster patterns: who is supported, at what ratio, in each half hour. Funding details are not identified, so the rosters show the pattern of support and not the extent to which any plan over-funds or under-funds it. These are designed rosters, built in a tool whose purpose is to resolve ratios, and they are likely to understate unfunded support in the wider sector. Figures are pooled across the rosters and rounded.
Exhibit 6 applies the first comparison to anonymised roster patterns from more than 100 rosters of care. The patterns record who is supported at what ratio in each half hour and do not identify plan funding. They show the pattern of support and the places where its ratios do not add up to whole workers. They cannot show how far any plan over-funds or under-funds its pattern, so plan overrun and unspent funding are not measured here.
Vacancy is the loss the sector discusses most. In these patterns unfunded support is the larger of the two, at about six percent of required hours against about five, and it sits in houses that have every place filled. Both are concentrated. The median house has neither. A quarter of houses have about a fifth or more of their required hours vacant or unfunded, which is more than the 12.7% a fully funded sleepover house earns, and a tenth have more than a third. These are designed rosters, built in a tool whose purpose is to resolve ratios, and we would expect the sector as a whole to look worse.
We do not have the second comparison for these houses. In our experience of matching payroll to required rosters, a house deviates from its required roster by about seven percent on average, in both directions. That figure is our estimate and is not drawn from the roster patterns. A house rostered above its required roster pays for hours that no plan funds. A house rostered below it is claiming for hours that were not worked, which is underservicing. The two gaps behave differently across a portfolio: rostering deviation runs in both directions and largely cancels, while vacancy and unfunded support only ever reduce revenue and so accumulate.
05The SIL house cost model
The losses described so far belong to the house, and the rest belong to the organisation. Together they fall into four groups, each measured against its own reference point, with two offsets. Table 1 defines each type.
In the roster of care. Vacancy, unfunded support and plan overrun: required workers that no plan pays for.
In rostering. Hours paid above the required roster, and overtime premiums. Hours paid below the required roster are underservicing, which shows as a gain. Most SIL houses have no paid handover period and rely on crossover shifts to pass information between workers. A house with one worker on shift at a time has no crossover, so handover is kept as a setting and is zero for the typical provider.
In revenue. Residents in hospital or away, and support that is delivered and never claimed or is claimed and rejected.
In the cost of an hour. Award levels, casual share and workers compensation that differ from the cost model, training and meetings beyond their allowance, and supervision and corporate overhead above theirs. The sector supervises at about one to twelve against the one to fifteen in the price.
Offsets. The sleepover surplus, and the break allowance. The price allows 4.17 percent of a worker's hours for breaks. Nobody leaves a SIL house for a break: the worker takes a paid working meal break and the house keeps billing, because covering a break in a one to one house would mean rostering a second person. Hours paid at SCHADS 2.1 instead of the 2.3 in the price also cost about six percent less.
Exhibit 7 puts the groups into one statement, in the form of the cost model in Exhibit 1. It takes each dollar of required roster at the price limit and divides it, in percentages, into the same layers of wages, operational overheads, corporate overheads and margin, with one layer added in front for the part of the roster that is never paid for. The first bar and column are a house with every hour funded, run as the cost model assumes. The second open on the average of the roster patterns: their mix of sleepover and awake nights, their vacancy and their unfunded support, with a typical provider's settings for everything else. Each layer opens to show its lines, and each line can be changed to describe any house.
Exhibit 7
The real SIL cost model
Each dollar of required roster at the price limit, divided into the layers of Exhibit 1, with the part that is never paid for in front. Opens on the average of the roster patterns. Select a layer to open its lines, and move any slider to describe any house.
The NDIA cost model: every hour funded, run as the model assumes
67%
14%
10%
8%
The real SIL cost model: this house
14%
66%
12%
12%
the dollar ends here
■ Not paid for, or not received■ Wages and on-costs■ Operational overheads■ Corporate overheads■ Margin
Margin on revenue, cost model
8.4%
Margin on revenue, this house
−4.5%
Vacant, unfunded or above plan, of the roster
11.0%
Revenue received, of the roster's value
86.3%
Select a layer to open or close its lines and sliders.
Layer
Cost model
This house
−Roster not paid for, and revenue not received
0%
13.7%
Vacancy
5.0% of the roster
0%
5.0%
Unfunded support
6.0% of the roster
0%
6.0%
Plan overrun
0.0% of the roster
0%
0%
Absence
1.00% of billing
0%
0.9%
Unclaimed revenue
2.00% of billing
0%
1.8%
margin on revenue with these settings: −4.5%
+Wages and on-costs for the required roster
67.1%
66.0%
+Operational overheads
14.0%
12.2%
+Corporate overheads
10.5%
12.0%
Margin
8.4%
−3.9%
Margin on revenue received
8.4%
−4.5%
What one step in each line is worth, in points of the roster's value
One paid active hour in each sleepover night
−2.20 pt
Ten more points of houses awake overnight
−1.03 pt
Supervision, one more point
−1.00 pt
Corporate overhead, one more point
−1.00 pt
Vacancy, one more point of the roster
−0.97 pt
Unfunded support, one more point of the roster
−0.97 pt
Plan overrun, one more point of the roster
−0.97 pt
Unclaimed revenue, one more point of billing
−0.88 pt
Absence, one more point of billing
−0.87 pt
Handover, five more minutes at each shift change
−0.87 pt
Incident and sanction loss, one more point of revenue
−0.86 pt
Non-billable time, one more point of worked hours
−0.72 pt
Workers compensation, one more point of wages
−0.69 pt
Payroll tax, one more point of wages
−0.69 pt
Rostered one point above the required roster
−0.68 pt
Overtime, one more point of hours
−0.44 pt
Ten points of hours moved from SCHADS 2.3 to 2.1
+0.41 pt
Ten points of hours moved from SCHADS 2.3 to 3.1
−0.36 pt
Casual share, ten more points of hours
−0.22 pt
Figures are percentages of the required roster valued at the price limit. The average takes three rounded figures from the roster patterns: about 35 percent of houses awake overnight, about five percent of the roster vacant and about six percent unfunded. Plan overrun is not measured in the roster patterns and opens at zero. Every other line opens at the typical provider's settings, which are our assumptions and not observations. The cost model column is the same night mix with every hour funded, run as the cost model assumes.
Two results stand out. The first is that the price limits leave the average house a margin of 8.4 percent when every hour is funded, and vacancy and unfunded support take about 11 percent of the roster. Run exactly as the cost model assumes, the average house loses 2.9%, and at a typical provider's settings it loses 4.5%. Plan overrun, which the roster patterns do not measure, would add to both. This is an average across houses and does not describe a typical house: most houses are fully funded, and the loss sits in the minority shown in Exhibit 6.
The second is the relative size of the two kinds of loss. A typical provider's operating settings, taken together, cost about 1.6 percentage points against the cost model: higher supervision, overhead, absence and unclaimed revenue, partly offset by the break allowance and the award mix. The roster of care costs more than six times that. The typical settings are our assumptions and not observations, and a provider should replace them with its own. The roster of care lines are the ones taken from data.
The lower part of the exhibit shows what one step in each line is worth. A point of vacancy, unfunded support or plan overrun costs about 1.0 percentage points, which is as much as a point of supervision or corporate overhead. Moving ten percent of hours from SCHADS 2.3 to 2.1 saves 0.4 points, and one paid active hour in each sleepover night costs 2.2 points.
Wages to revenue
Most providers watch one figure for a house, which is support worker wages as a share of revenue. The model gives that figure a reference range, and Exhibit 8 calculates it from a small set of assumptions. Wages here are everything paid to the support workers on the roster, with leave, superannuation and allowances, including overtime and paid time that is not billed. Supervision is not included.
Exhibit 8
Wages to revenue
Support worker wages, with leave, superannuation and allowances, as a share of the revenue the house receives. Overtime and paid time that is not billed are included. Supervision is not.
Nights
Hours paid at SCHADS 2.1, with the rest at 2.325%
Casual share of hours25%
Paid time that is not billed, of worked hours2.50%
Hours paid as overtime2.0%
Roster that no plan pays for11.0%
Billing that is not received3.00%
Wages to revenue, this house
79.4%
With workers compensation
81.9%
Lowest under optimal conditions
63.9%
Highest under optimal conditions
73.4%
This house above its own optimal
11.7 pt
this house 79.4%
60%
70%
80%
90%
■ range under optimal conditions, sleepover house to awake house, at this workforce
Under optimal conditions
Workforce
Lowest: sleepover house
Highest: awake overnight
As the cost model assumes
68.7%
79.1%
A typical provider's workforce
63.9%
73.4%
The workforce set above
63.9%
73.4%
Optimal conditions are every rostered hour funded, claimed and paid, with no overtime, at the workforce set above. The lowest figure is a house covered by a sleepover every night and the highest is a house staffed awake overnight. A house with more than one worker on shift by day sits between them. Workers compensation is at the cost model's rate.
Under optimal conditions, with every rostered hour funded, claimed and paid, the ratio is set by the shape of the roster and by the workforce. Run as the cost model assumes, a house covered by a sleepover has a ratio of 68.7% and a house staffed awake overnight has 79.1%. The sleepover house is the lowest, because the sleepover is paid as an allowance worth about a fifth of its price. The awake house is the highest, because every hour carries a worker who costs about 72 percent of the price before any non-billable time. With a typical provider's workforce the range is 63.9% to 73.4%. The 63.2% shown for the sleepover house in Exhibit 2 is the wage for the rostered hours alone, before paid non-billable time.
The range is a test. A house whose ratio sits above the top of the range for its workforce has hours on its roster that are not being paid for, or is paying for them at overtime. The average house in the roster patterns, at a typical provider's settings, has a ratio of 79.4%, which is 11.7 points above the 67.7% it would show if every hour were funded and received. The ratio also has a break-even point. A fully funded house breaks even at about 81% at the cost model's overheads, and at about 78% with a typical provider's supervision and corporate overhead.
The operating settings in Exhibit 7 carry into Exhibits 9 and 10. Exhibit 8 has its own.
06A provider's result is a distribution of houses
A provider operates a set of houses, and its result is the sum of their differences. The portfolios in Exhibit 9 are synthetic. About a third of the houses are staffed awake overnight, plan budgets vary around their rosters by an assumed amount, we assume that four houses in five have plans that step up, and each house deviates from its required roster by seven percent on average in either direction. The portfolio margin conceals a wide spread between houses, and a minority with vacant places, unfunded support or plan overruns account for most of the loss.
Exhibit 9
A portfolio of forty houses
Each dot is a house: the share of its rostered hours that is vacant or unfunded, against its margin. The houses are run at the operating settings chosen in Exhibit 7.
Houses
Operating settings
At these operating settings the forty houses would earn about 8.2% if every rostered hour were paid for. The vacant places, unfunded hours and plan overruns in the houses take 12.5 points, which leaves −4.3%. Those losses belong to the houses in the portfolio selected here. The vacancy, unfunded support and plan overrun sliders in Exhibit 7 describe a single house and do not carry into this exhibit.
Portfolio margin
−4.3%
Margin if every rostered hour were paid for
8.2%
Houses with a negative margin
25 of 40
Rostered hours vacant or unfunded
8.2%
Billing not received
6.3%
Revenue a year
$29.8m
Rostered against the required roster, house by house
House to house rostering deviation, either way7.0%
Average deviation, either way
6.0%
Houses rostered below required
18 of 40
Paid below the required roster, a year
$658,538
Paid above the required roster, a year
$665,982
ILLUSTRATIVE. These portfolios are constructed to exercise the model and are not observed data.
07Pooling, and the claim that ratios do not exist
Pooling treats the plans in a house as one budget. Each resident's claim is set by what their plan can pay, and not by the ratio at which the support was shared, so that the plans together cover the roster. Our position is that it should not be used.
Residents who share a worker would ordinarily be claimed at the same ratio for the hours they share. Under pooling the ratios are fitted to the plans. A half, a quarter and a quarter looks reasonable and adds up to one worker. It is a cross-subsidy: three people are sharing one worker and one of them is paying for half of it. Nothing in the method limits the split, and ninety-nine hundredths and one hundredth also add up to one worker. The ratios that make a set of plans cover a roster can be solved for directly. For the three plans in Exhibit 4 they are one ninth, one third and five ninths. At those ratios the house moves from a loss of 7.5% to a margin of 11.2%, and the resident with the $500,000 plan pays about $139,000 a year for support delivered to housemates.
Exhibit 10
Three plans treated as one budget
The house in Exhibit 4: three residents sharing one worker on a sleepover, with plans of $100,000, $300,000 and $500,000, at the provider settings above.
Approach
Workers rostered
Workers claimed
Revenue a year
Margin
Paid for housemates
Claimed, not rostered
Each resident is claimed a third of the worker they share
1
1
$502k
−7.5%
0
0
Pooled: each claim is fitted to its plan so that the three plans cover the worker (0.11, 0.33, 0.56 of a worker)
1
1
$607k
11.2%
$139k
0
Each plan's weekly amount is claimed in full, with one worker rostered (0.16, 0.48, 0.80 of a worker)
1
1.44
$873k
38.2%
0
$274k
Paid for housemates is what a resident pays beyond an equal share of the worker the house shares. Claimed, not rostered is the value of claims for fractions of a worker beyond the workers on the roster.
Pooling and underservicing are different things
Underservicing is rostering fewer workers than the roster of care requires. It can happen in any house and does not need pooling. Pooling is a claiming practice, and a pooled house can be rostered exactly as required, as it is in the second row of Exhibit 9.
SIL can be claimed by the hour or as a weekly amount, which is the participant's regular SIL funding divided by the weeks in the plan. Weekly claiming requires a service agreement that records the agreement to claim weekly and the agreed typical schedule of supports, and the Agency's guideline is that a provider can charge only for services that were delivered and agreed. A weekly claim is therefore tied to a schedule of supports, and the schedule is what the claim has to be reconciled to.
Pooling and underservicing are connected because pooling sets the weekly claim from the plans and leaves the schedule behind. Once the claim no longer comes from the roster, nothing ties it to the labour delivered, and a provider can roster below the required amount without any change in what it claims. That is how pooling tends to lead to underservicing. In the third row each plan's weekly amount is claimed in full, the house claims 1.44 workers against the one it rosters, and the margin rises to 38.2%. At that point the margin is set by the size of the plans in the house and not by the cost of the labour. The price limit is built as cost plus two percent, and the claiming rules require a claim to match the supports that were provided.
Why we advise against it
It overrides the Agency's assessment. The Agency's guideline is that SIL funding depends on the participant's own disability support needs, and not on who else shares their supports, and the Agency uses the support ratio as its measure of the level of support a participant needs. Pooling moves that support between housemates. It is in effect a reassessment made by the provider alone: the participant assessed at the higher ratio receives less than was assessed, and the housemates receive more than was assessed for them.
The provider cannot show that a participant received what they paid for. A resident claimed five ninths of a shared worker would have to be shown to have received five ninths of that worker's time, which would take a record kept minute by minute. The claiming rules allow a claim only after a support has been delivered and require it to match what was provided. Where a worker supports several participants at once, each participant's price limit is the hourly limit divided by the number sharing, and a worker's time can be claimed against a participant only while the worker is supporting that participant. In effect the housemate claimed a ninth is receiving on-call support they have not paid for.
It conceals what the participant is paying for. A schedule of supports that set out a pooled arrangement honestly would show ratios that no reader could interpret, such as five ninths of a worker. In practice the schedule either leaves them out or cannot be checked against anything, so the participant and their guardian cannot see what is being paid for or scrutinise it. The Agency's guidance to SIL providers is that the provider and the participant should agree on the roster of care and on how the plan funding will be claimed. Its guideline says a service agreement should cover the ratio and intensity of support and its price, and tells participants they can ask their provider for their roster of care at any time. The NDIS Practice Standards require that each participant understands the supports they have chosen and how they will be provided. A pooled arrangement meets none of these, and it is contrary to the choice and control that the NDIS Act sets out as a general principle.
It works against the participant with the larger plan. Support moves from the resident funded at the higher ratio to the residents funded at lower ratios, and none of them can see it. At that participant's next plan review, the provider's own roster is the evidence that they were supported without the ratio they were funded for.
It is potentially unsafe. The Agency's guideline funds one to one support around the clock where a participant has life-threatening or constantly monitored health needs, frequent high-risk behaviours of concern, or a condition that is changing quickly. The NDIS Review's position was that 24-hour living supports would otherwise generally be funded at one to three, and the funding changes of October 2026 use 24-hour support at one to three as their benchmark. A ratio more intensive than one to three is therefore funded where the participant's need for it is understood and documented. Reducing that participant's support without disclosure, and using the difference as a subsidy for the provider's convenience, puts at risk the resident whose need was assessed as greatest.
It conceals the cost of support. A pooled house shows plans that are fully used and no shortfall, so the Agency receives no signal that a plan was too small, that a ratio did not work, or that a placement was poorly matched.
It does not survive a reconciliation. The difference between a pooled house and a compliant one shows when the roster that was worked is compared with the ratios that were claimed. The Agency's record-keeping guidance requires providers to keep staff rosters and says that a roster for shared supports should show the ratio of support, and since 2026 the Agency has had the power to require evidence before a claim is paid. Our observation is that it is willing and able to use that power to take a roster apart, comparing the support that was rostered and paid for with the support that was claimed. Under commissioning the same question will come from a purchaser who holds far more information than a claims system does.
The argument made for pooling is that ratios do not exist in a home: support is shared fluidly, needs vary through the day, and a funded ratio is only the method the Agency used to arrive at a budget. Taken to its conclusion the argument dispenses with the roster of care, so that no schedule of supports is declared to the participant and the weekly amount is claimed from each plan. If the argument is right, the provider has no basis for the amount it charges each plan and nothing to reconcile when it is asked.
The financial incentive nonetheless runs toward pooling, and it is strongest in the houses described in sections 02 and 03, where the plans together hold more than the roster is worth and the house still loses money. A provider that keeps residents on the ratios they share carries unfunded support and plan overruns as a cost and reports the lower margin. A later paper deals with these incentives and what they imply for commissioning.
Pooling is strictly against the interests of the higher funded participant. It violates their choice and control, and it does so surreptitiously, in a form that neither they nor their guardian can see. Reviewed on that basis, we would expect the NDIA to take a dim view of the practice, and we ask that it does.
08What we advise
Construct the roster of care with the residents and their representatives, and keep it current. It is the record of what each plan is paying for.
Keep residents who share a worker on the same ratio for the hours they share. Decide for each part of the week who is sharing, and claim each plan only for the fraction its participant receives.
Record each loss under its own name. Where the roster of care leaves funding unspent, leave it unspent. Where it leaves a shortfall, decide whether it is carried as unfunded support or as a plan overrun, record it as that, and seek a plan review with the roster of care as the evidence.
Examine the roster of care before filling a vacancy. A vacancy is expensive: the vacant place in Exhibit 2 leaves about $209,000 a year unfunded. Filling it with a resident whose plan is $100,000 leaves a plan overrun of about $109,000. That is a smaller loss than the vacancy, and it lasts for as long as the placement does. A placement accepted under vacancy pressure without this examination can leave a house in a position that cannot be resolved within the rules.
Reconcile the roster of care to payroll regularly. Do it before someone else asks for the reconciliation.
Providers are under pressure to do otherwise. A vacant place leaves its share of a worker unpaid from the day it opens. The Agency does not fund vacancies, and its guideline is that a provider should not take extra money from the other residents' plans to cover one. A provider has to fill the place quickly to stay solvent, and in most regions few new residents are looking for a place. Vacancy is the subject of the next paper in this series.
09What follows
This paper is the reference for a series, and the later papers use its model and its definitions without restating them. They cover:
who captures the value of a vacancy, and why participant choice cannot be what fills one;
how the incentives in section 07 behave when quality cannot be observed and the price is fixed;
why payments for outcomes do not work in a service where the provider does not choose the cohort and the counterfactual cannot be observed;
what a commissioner would need to collect to tell a compliant house from a pooled one.
Empathia Insider
Sector analysis, in your inbox.
Occasional, considered analysis on SIL economics, NDIS reform, and provider viability. And you can unsubscribe anytime.
10The leakage types, defined
A vacancy, unfunded support and a plan overrun are the same in their effect on a house. The purpose of separate definitions is to let a provider say which one it has, where it is recorded and who can fix it. The same applies to the rostering, revenue and cost types.
Table 1
The leakage types, defined
Each type is measured against a different reference point, which is what allows one to be told from another.
Type
What it is
How it is measured
In the roster of care: required less funded
Vacancy obligation
The share of a required worker that is attached to a place with no resident in it.
Required less funded, on the service pattern held for the vacant place.
Unfunded support
Required workers beyond what the residents' ratios add up to, with every place filled. It arises where funded ratios do not sum to whole workers, and where the group sharing a worker changes and the plans do not step up.
Required less funded, after the vacancy obligation is taken out.
Plan overrun
Support rostered and claimed at a ratio the participant's plan cannot pay for over the plan period. It has the same effect as unfunded support and appears against the plan, not on the roster. Under weekly claiming it appears as an agreed schedule that costs more than the weekly amount.
The roster of care priced over the plan period, less the plan budget, where that is positive.
Unspent funding
Plan funding that the roster of care does not use. It is not a cost to the provider. It is a risk to the participant at the next plan review. Under weekly claiming it does not appear, because the weekly amount is claimed in full, and the comparison has to be made between the weekly amount and the schedule delivered.
The plan budget less the roster of care priced over the plan period, where that is positive.
In rostering: rostered less required
Over-rostering
Paid hours above the required roster, such as added shifts, extended shifts and duplicated cover.
Rostered less required, where positive.
Underservicing
Paid hours below the required roster while claims continue to follow the roster of care. It appears as a gain in the accounts and is a compliance exposure.
Rostered less required, where negative.
Handover overlap
Two workers paid for the same minutes at a change of shift. Most SIL houses have no handover period and rely on crossover shifts, so it arises mainly where one worker is on shift at a time.
Paid overlap at each shift change.
Overtime
The premium paid over the ordinary rate for hours worked as overtime.
Overtime premium as a share of rostered wages.
In revenue: received less claimable
Absence
Days when a resident is in hospital or away, the roster is held and the support cannot be claimed.
The roster of care's billing for days of absence.
Unclaimed revenue
Support that was delivered and never claimed, or claimed and not paid.
Claimable billing less payments received.
In the cost of an hour: actual less the allowance in the price
Wage and on-cost variance
The award levels paid, the casual share, the workers compensation premium and payroll tax, where they differ from the cost model.
The loaded cost of an hour less the cost model's.
Non-billable time
Training, meetings and any breaks that stop billing.
Paid non-billable hours less the allowance in the price.
Supervision
Supervisors' cost above the price's allowance of one supervisor to fifteen full-time equivalent workers.
Supervision cost less the allowance.
Corporate overhead
Overhead above the allowance in the price.
Overhead cost less the allowance.
Offsets
Sleepover surplus
The difference between the sleepover price and the cost of a sleepover shift.
Sleepover billing less sleepover cost.
Break allowance retained
The allowance for breaks in the price, where workers take a paid working meal break and the house keeps billing.
The allowance less breaks that stop billing.
11Method, data and limits
Prices are the 2026-27 Pricing Schedule limits for standard intensity SIL at national rates. Wages follow the NDIA cost model for DSW Level 1 at SCHADS 2.3, with leave of 21.7% of base pay implied by the model's cumulative on-cost of 37.51 percent, and an award mix across levels 2.1, 2.3 and 3.1. The typical provider has no hours at 3.1, which belongs to higher intensity houses whose price limits are not modelled. A week has 70 weekday daytime hours, 10 weekday evening hours, 16 daytime hours on each weekend day and 56 night hours. A house is either staffed awake overnight at its daytime ratios or covered by a sleepover, and mixed arrangements are not modelled. Public holidays are costed as eleven days a year on the Sunday roster. Workers are rounded up within each funded ratio, and each plan pays only for its own participant, except in Exhibit 10. Supervision and corporate overhead are costed on the workers rostered, valued at the price limit, as they are in the cost model, and do not fall when a place is vacant or an hour is unfunded. A full, whole, awake house at the cost model's settings reproduces the cost model's two percent margin.
The roster data are anonymised roster patterns from more than 100 designed rosters of care held in Empathia's RoC Optimiser. They carry no funding details, and they are not a sample of the sector. Every figure that involves a plan budget in this paper comes from the model and its stated examples, and none comes from the roster data. Exhibit 7 applies the vacant and unfunded shares of required hours as shares of the roster's value, and applies the share of houses awake overnight to houses of equal size. The rostering deviation in section 04 is Empathia's estimate from matching payroll to rosters of care for providers, and is not a measured result. Revenue is modelled as the value of the support delivered to each participant, up to that participant's funding. That is what an hourly claim produces and what a weekly claim should reconcile to. Three limits should be kept in mind. The paid active time within a sleepover night is set to zero. It has a large effect on the margin of sleepover houses and is a setting in Exhibit 7. High intensity price limits are not modelled. The portfolios in Exhibit 9 are constructed and are not observed houses.
Disclosure. Empathia's principals hold interests in NDIS providers and are engaged by organisations acquiring them. Empathia builds and sells the RoC Optimiser and the SIL Dashboard, the software referred to in this paper.
12Sources
NDIA, Disability Support Worker Cost Model 2026-27, and Disability Support Worker Cost Model 2021-22 (parameters of operational overheads).