Noesa
← The Casebook

Australia · 2015–2023

It spread a year's pay evenly across the fortnights, then called the difference a debt.

At least $1.763 billion raised against 433,000 people, calculated from a fortnightly income none of them had actually earned.

Published 18 August 2026

What happened

Between July 2015 and November 2019 the Australian Government ran an automated debt-recovery system against people who had received social security payments. It took a recipient's annual PAYG income as reported to the Tax Office, apportioned that income evenly across fortnightly increments of the review period to produce what the parties called a "notional" or "assumed" fortnightly income, and compared it with the actual fortnightly income the person had reported to Centrelink — the figure their payments had originally been assessed on. Where the notional figure was higher, the Commonwealth treated the difference as an overpayment and raised a debt [1].

The Commonwealth pursued repayment, including through private debt collection agencies [1]. In December 2016, as the scheme was being ramped up, the then-Minister for Human Services Alan Tudge said on national television: "We will find you, we will track you down, and you will have to repay those debts, and you may end up in prison." The Royal Commission later recorded that people receiving income support had been made to feel "shame, oppression, isolation, and dehumanisation" [2].

The warnings arrived early. In April 2017 the Commonwealth Ombudsman reported on the system, finding "deficiencies in (the department's) service delivery and communication to customers and staff when implementing the system", that debt letters "were unclear and did not include crucial information", and that the 21-day window given to respond "was not reasonable or fair in all circumstances"; the Ombudsman also found that around one in five people sent debt notices did not owe any money [3]. The Royal Commission would later put it more bluntly — that "the beginning of 2017 was the point at which Robodebt's unfairness, probable illegality and cruelty became apparent", and that instead of stopping, the path taken "was to double down, to go on the attack in the media against those who complained and to maintain the falsehood that in fact the system had not changed at all". Review volumes then grew rather than shrank: 681,292 reviews from early 2017 under one programme and 247,459 from October 2018 under another, against 220,439 before 2017 [2].

The scheme was broken by an individual case. Deanna Amato, a local government employee, was told she owed $3,215.38. Centrelink had sent correspondence to an address she had left almost three years earlier, and took her tax refund of more than $1,700 without her knowledge; she had worked full-time for one employer and then moved to casual work when she began studying, so averaging overstated her income roughly fourfold for the fortnights in question. On 27 November 2019 the Federal Court, by consent, declared that the demand for payment was not validly made because the information before the decision-maker was not capable of satisfying them that a debt was owed. The Commonwealth conceded that raising a debt solely on averaged Tax Office data was unlawful. Her actual overpayment was $1.48, and a later freedom-of-information request showed she had in fact been underpaid by just over $480 [4].

A class action followed. On 11 June 2021 Justice Murphy of the Federal Court approved its settlement, recording that the Commonwealth had admitted during the proceeding that it had no proper legal basis to raise, demand or recover debts based on income averaging, and that on the evidence it had "unlawfully asserted such debts, totalling at least $1.763 billion against approximately 433,000 Australians", recovering approximately $751 million from about 381,000 of them. The Commonwealth agreed — without admission of liability — to pay $112 million inclusive of legal costs, not to pursue the invalid debts, and to consent to declarations that decisions resting solely on averaged Tax Office data, absent other evidence the person had earned at a constant fortnightly rate, were not validly made. By then it had already refunded $707.9 million. Justice Murphy called the proceeding "a shameful chapter in the administration of the Commonwealth social security system and a massive failure of public administration". Of the roughly 648,000 group members, some 394,000 stood to recover; objections filed with the Court described financial hardship and distress, some group members wrote of contemplating suicide, and one bereaved mother told the Court her son had died by suicide after the Commonwealth demanded payment of a debt she says he did not owe [1].

A Royal Commission delivered its report to the Governor-General on 7 July 2023, and it was tabled in Parliament the same day [2]. Commissioner Catherine Holmes AC SC found that "Robodebt was a crude and cruel mechanism, neither fair nor legal, and it made many people feel like criminals", that "people were traumatised on the off-chance they might owe money", and that it was "a costly failure of public administration, in both human and economic terms" [5]. She made 56 recommendations, directed principally at strengthening the public service and the oversight agencies that had failed to stop it; in November 2023 the Government accepted or accepted in principle all 56 [2].

Where the same matching earns its place

Comparing tax records against declared income is legitimate, and no court in this saga said otherwise. The same judgment that condemned the scheme drew the line precisely. A large category of people had debts that were not based on averaging at all: their debts had been "assessed by the Commonwealth from payslips, bank statements and other information they provided", and Justice Murphy considered legal claims against those debts weak, and more likely than not to fail at trial [1]. Same data match, same agency, same years — the difference was whether a discrepancy opened an investigation or closed one. Deanna Amato's file is the demonstration in miniature: once someone actually examined what she had earned fortnight by fortnight, $3,215.38 became $1.48 [4].

Where it burned

The arithmetic never failed; the quantity it started from did not exist. Income support is assessed on what someone earned in each particular fortnight, and recipients' work is commonly casual, part-time, sessional or intermittent — so a year's earnings divided evenly across fortnights describes nobody's actual fortnights. Justice Murphy wrote that it "should have been plain" that the system "may indicate an overpayment of social security benefits when that was not in fact the case" [1]. The defining move was what the scheme did with that gap: rather than treat it as a question to investigate, it raised a debt, and "in the absence of further information from social security recipients" the assumption stood as the finding [1]. Recovery followed, private debt collectors included [1], while the person had 21 days the Ombudsman called "not reasonable or fair in all circumstances" to assemble payslips from years earlier [3] — letters that might have gone to an address they had long since left [4].

The tell

When a number is produced by spreading a total evenly across a period, ask whether the rule it feeds depends on the timing inside that period — because if it does, the average has erased the only thing that mattered, and no amount of correct arithmetic afterwards puts it back.

Averaging is not an error. Applying an average to a rule that is defined on the variation is, and that is a single, small, entirely legible mistake — which is precisely why it survived more than four years and reached hundreds of thousands of people. What made it ruinous was the second half, and the second half is the part that generalises: the output was issued as a debt rather than as a question, so the cost of the system being wrong moved onto the person least equipped to carry it, and its accuracy stopped being the agency's problem. You will meet this shape well outside welfare — fraud scores, risk tiers, usage-based billing, insurance pricing, performance ratings all smooth something away and then bill you for the residue. So carry two questions rather than one: what did the average smooth over, and who is being made to pay for finding out it was wrong?

Share this case

The image has the link printed on it, so it still leads back here.

The check is a habit, and habits are trained. Statistics, understood teaches the exact move that failed here — that an average is a statement about a distribution and not about any member of it, and that the moment a rule depends on variation, replacing the distribution with its mean discards the evidence the rule runs on. Robodebt is what that substitution looks like at national scale, and the course's habit of asking what a summary statistic destroyed is the one that catches it.

Sources

Every source below was opened and read. Last verified 17 August 2026.

  1. [1] Prygodicz v Commonwealth of Australia (No 2) [2021] FCA 634 — Summary issued with the orders (8 pp.)Federal Court of Australia (Murphy J), VID 1252/2019, 11 June 2021
  2. [2] Government Response | Royal Commission into the Robodebt SchemeAustralian Government, Department of the Prime Minister and Cabinet, November 2023
  3. [3] Centrelink's debt recovery system lacks transparency and treated some customers unfairly, ombudsman findsABC News (Australian Broadcasting Corporation), reporting the Commonwealth Ombudsman's April 2017 report, 10 April 2017
  4. [4] An in-depth look at our robo-debt test caseVictoria Legal Aid (solicitors for the applicant in Amato v Commonwealth)
  5. [5] Final report of the Royal Commission into the Robodebt SchemePrime Minister of Australia, 7 July 2023