Cross-register provider integrity

The problem

The National Disability Insurance Scheme and the aged care system pay many of the same operators. Both record the operator’s ABN. Neither register tells the other when a provider’s status changes.

So a registration can expire, or a provider can be suspended after enforcement action, while the same operator holds an active approval to deliver high-value supports in the other system. Nobody has mishandled either register. The gap is between them, and no single regulator owns it — which is why it persists.

None of this needs new powers, a new register, or a change to how either regulator works. It needs someone to read across both, continuously.

How it works

Everything here runs on published sources: the disability and aged care provider registers, regulatory actions and banning orders, corporate records, the charity register and the business register.

Entity resolution. We match on ABN first, then resolve the same operator across differing legal names, trading names and register names, including trusts that carry no public director record. This is the step that produces value nothing else currently delivers: not a list of providers, but one operator seen whole.

Scoring. Each typology has a written rubric with weighted components — name similarity to a banned or sanctioned provider, a shared address with a similar-named provider, matching support categories in the same state, and the registration and cancellation history on the ABN. Every flag shows which components produced its score, and at what weight. We publish the method to the agency that engages us, not to the internet, because a rubric on a public page is a rubric to be gamed.

Collection. Agents run to a schedule against public sources. A cancelled ABN, a banning order or a lapsed registration surfaces at the next run rather than when somebody reports it.

Triage. An analyst reads each flag, works through its components, checks the corroborating public material, and decides whether it is worth an investigator’s time. The referral carries that decision and the name of the person who made it.

What a pass over public registers found

We matched 2,169 registered aged care providers against more than 23,000 registered disability providers by ABN, fused the result with corporate and charity records, then checked the flagged entities against open sources.

Twenty-four providers with aged care enforcement action against them matched to the disability register. Three were still active there.

• An expired aged care registration alongside an active disability approval covering 21 registration groups, including the highest-value accommodation supports. Three director changes were filed in the three months before the registration lapsed.
• One ABN trading under three names across three registers, with a trust as the legal entity and no public director record. The business was still advertising for staff.
• A provider suspended from aged care still holding a plan management registration — control of participant funds without delivering a service.
• Twenty providers at a single address cluster, carrying two active banning orders and eight disqualified directors between them. One had a cancelled ABN and was still listed across 17 support categories.

The match that produced these took about two seconds of compute against more than 25,000 provider records. It has never been run as a routine compliance check.

We do not publish the names. Nothing has been proven against any operator behind these patterns, and naming them here would make us the accuser — the role this service deliberately does not take. The names, with their sources and the steps that produced them, go to the agency that can act on them.

What the agency provides

For a proof of concept: nothing. No data-sharing agreement, no systems access, no extract. Every finding above came from public sources, so we can run the method against your provider population and show you the result before you give us anything. That removes the slowest step in a government engagement.

For continuing operation:

• A named contact who can tell us whether a referral was useful.
• A route for referrals into the investigation workflow you already use.
• Program data, under an agreed data-sharing arrangement, if you want detection to reach past what is public.
• Outcomes, so the scoring learns from what your investigators actually found.

What we deliver

• A ranked referral list. Each referral carries its confidence, the components that produced the score, the corroborating public material, and a specific next step: which extract to buy, which regulator to ask, which discrepancy to reconcile.
• Entity views — aliases, related parties, address clusters — with the register each fact came from.
• A standing picture, refreshed on the collection schedule, rather than a report that is stale the week after it lands.
• The assurance record: bias testing, model versions, and the decision trail behind every referral.
• An export your case management system can read.

How it is measured

• Referrals accepted by your investigators, as a share of referrals made. This is the measure we expect to be judged on.
• Cost per accepted referral.
• Time from a signal appearing in a public source to a referral reaching you.
• Disparity ratios across the six tested dimensions, with review at 1.25 and a stop at 1.5.
• Precision and recall, per model version, reported whether they improved or not.

These are also the measures the performance component is settled against, so the reporting and the commercial arrangement run off one set of numbers rather than two.

We do not report value recovered. We have not measured it on this capability, and a number we cannot evidence is worse than no number.

How it is priced

Three parts. The proof of concept is a fixed fee. The base subscription keeps the capability running. The third moves with what the capability does for your program.

• Proof of concept. A fixed fee for a fixed window and one defined question, quoted after a scoping conversation.
• Base subscription. An annual fee, sized to the provider population and the collection cadence, covering collection, fusion, scoring, analyst triage and the assurance record. Quoted once the population and cadence are agreed.
• Performance component. A share of the contract value settled against a basket of service metrics agreed with you before the work starts. The share is agreed with you and written into the contract.

Why part of the fee should move

A base fee buys continuity, not improvement. It pays for the capability to keep running at the standard it runs at today. Nothing in it rewards us for making it better, and nothing in it costs us if it stops being useful to you. A flat fee is the right price for a service that should not change. It is the wrong price for one that has to keep up with people who change what they are doing as soon as they are caught.

It puts you in a risk and reward position rather than only a cost position. Public money generally has to be justified before the benefit exists, on an estimate of what it will return, and then tested long afterwards by a review, an audit or a committee. That cycle corrects more than it improves: little has to change until somebody finds a problem, and by then the money is spent. A component that moves with agreed measures brings the judgement forward. You pay it in full when the measures move, you pay less when they do not, and you can see which one is happening each quarter rather than at the end of a term.

It decides where our attention goes. Every supplier drifts towards whatever grows its revenue. If our revenue can only grow when your measures move, that drift runs towards your program instead of away from it — and it keeps running that way in year three, when the novelty has gone and the original team has moved on. This is the difference between a commitment we assert and one you can enforce.

The measures

The basket is yours. We do not arrive with our own scorecard. It is drawn from the measures your program is already accountable for, weighted as you weight them, and written into the contract:

Two of these turn on decisions only you can make. Whether a referral is assessed, and whether an action follows, depends on your assessors and their workload, not on us. So they are set as shared measures: if assessment capacity moves to something more urgent for a period, the measure is adjusted rather than failed. An incentive that penalises us for your resourcing — or that quietly rewards you for acting on less — is a dispute waiting to happen, and it should be settled in the contract rather than in the first quarter it bites.

Better detection also makes the numbers go up before they go down. More looking finds more, so a typology measure read over one quarter would punish us for working. It is read over a year, against a baseline agreed before we start.

How the money is held

This is the part we would rather raise than leave for later. Money that might be paid still has to be committed when the contract is signed, so it sits against your budget whether or not it is earned. Set it aside and an unearned component becomes an underspend somebody has to explain. Do not set it aside and an earned one arrives where there is no headroom. Neither is a reason to pay a flat fee, but both are real, and the mechanism belongs in the contract rather than in an awkward conversation at the end of the first year.

Three arrangements that work, depending on how your finance area prefers to carry a variable commitment:

• At risk inside the full fee. The contract is written at its maximum and the invoice reduces when a measure is missed. Nothing new has to be found to pay us, and of the two failure modes you get the underspend.
• A performance pool inside an approved ceiling. The maximum is approved once; the split between base and performance is settled each period against the basket.
• Performance decides the option year. No money moves between periods at all. What performance buys is whether the extension is exercised.

The first is usually the easiest to approve, because it never asks for money that was not already committed. We will take whichever treatment your finance and procurement areas can sign off fastest. The mechanism matters far less than that the money follows the measures.

How it is kept honest

An incentive paid on the wrong measure buys the wrong behaviour, so:

• You measure and report the basket. We do not mark our own homework, and we will accept your figure over ours where they differ.
• Nothing pays on volume alone. A referral your assessors reject counts against the accepted-referral measure, not for it, so sending more of everything is the worst move available to us.
• Bias testing is a gate, not a scorer. If a model crosses a 1.5 disparity ratio in a period, no incentive is payable for that period regardless of the other measures.
• The component is capped, and the base subscription is set so the capability keeps running whether or not any of it is earned. The cap is agreed with you before the work starts.
• Handover milestones sit in the basket alongside the rest. We are paid partly on making ourselves replaceable, so the exit stops being a threat to our revenue.
• The basket is reset each year. A measure left in place long enough stops being a measure and becomes a target to farm.

What we will not take. No share of recoveries and no fee per referral. Both pay us to push more work at your investigators whatever its quality, and both would put a commercial interest inside a referral that has to survive review.

What it does not cover

• Claims data. The capability runs on public registers and published data. It holds no MBS, PBS, NDIS payment or aged care claims data.
• Billing typologies. Over-servicing, upcoding, duplicate claims and phantom billing are detected from claims data. They are outside this capability.
• Real-time monitoring. Collection runs to a schedule. A change in a public source appears at the next scheduled run.
• Determinations. We do not decide whether fraud has occurred, and we do not make or recommend a statutory decision. Every output is a lead for your assessors.
• Investigation. We do not contact providers, participants or staff, and we do not gather evidence. Analytical support to your investigators during a case — further entity work, framing a data request, walking an assessor through how a flag was produced — is available at a daily rate, scoped and quoted before it starts.
• Other programs. This entry covers the disability and aged care provider populations. Procurement, grants and state-funded services sit on different registers.

Each of these can be scoped as separate work. None is included in the fee above.