“Robodebt” is the label applied to an Australian government initiative starting in 2016 designed to increase recoveries of “overpayments” made to social security recipients, retrospectively dating back to 2010. The original objective was to recover $1.7 billion over five years from “overpayments” to current and former social security recipients.
Robodebt involved data-matching historic records of benefit payments made to individuals with past income tax returns, identifying discrepancies between these records. It reduced human investigation of the discrepancies, with the automatic calculation of overpayments for many individuals based on a simple algorithm that averaged earnings over the relevant year, followed by raising of a debt against those individuals. The use of this “algorithm” led to the initiative being labelled on social media and in mainstream media as “Robodebt”.
From late 2016, there were extensive complaints from affected individuals and negative media reporting. The controversy escalated, with the shadow human services minister requesting the auditor general to investigate, and an independent MP separately asking the Commonwealth ombudsman to investigate after receiving more than 100 complaints about the debts.
In January 2017, a website now sharing more than 1,200 personal accounts, a Facebook account and a Twitter account with the hashtag #notmydebt were all set up to record contested debts. A number of other individuals, journalists and legal academics consistently highlighted the issue on social and mainstream media.
There were subsequently two reports by the Commonwealth ombudsman and two parliamentary committee inquiries (one ongoing). Despite this high level of scrutiny, the scheme only started to unravel in 2019, due to cases brought by Victoria Legal Aid. In 2018, Centrelink raised a debt of over $3,700 against a former student, Madeleine Masterton, and in February 2019, VLA filed an application for judicial review. Within a week, the debt was reduced to around $600. At the first case management hearing, Centrelink accepted Masterton’s original declared income, with the result that there was no debt, and no legal ruling on the issue.
The second case also involved a former student, Deanna Amato, who only became aware that a debt of just over $3,200 had been raised against her — plus a penalty of 10% for “not engaging” — when her income tax refund of around $1,700 was taken as repayment. After the litigation began, Centrelink gathered information from her employers to determine the debt should be reduced to $1.48. (A subsequent Freedom of Information request revealed she had actually been underpaid by $480!)
In September 2019, the Department of Human Services (DHS) completely dropped the Amato debt but refused to pay interest. Separately, the Federal Opposition facilitated a class action. In November, the Federal Court ruled that income averaging was unlawful – a conclusion the government conceded a week before the case came to trial. This ruling also required the government to have more than 500,000 individual debts manually recalculated.
The class action bought by Gordon Legal was settled in November 2020 at a cost to the government of around $1.2 billion. This comprised refunds of $721 million to 373,000 people, $112 million in compensation and $398 million in cancelled debts. While this settlement is reported as being the largest class action settlement in Australian history, this may not be the full cost to the Budget as it does not include the administrative costs involved in the recalculation of “debts”.
The proposed settlement was approved by the Federal Court on 11 June 2021, but the cost has increased to more than $1.8 billion, including interest payments of $112 million. Justice Murphy of the Federal Court condemned the Government’s actions in the clearest terms: “The proceeding has exposed a shameful chapter in the administration of the commonwealth social security system and a massive failure of public administration.”
There have been calls by Opposition parties for a Royal Commission into the programme. Submissions to Parliamentary Inquiries and media reporting document the personal suffering of people affected, including cases where families claimed that affected individuals committed suicide. Around 680 people who objected to the settlement will be allowed to opt-out – and may pursue further claims, as some believed the interest payments were insufficient and the government had not been held accountable for its mistakes.
What went wrong?
Data-matching with the Australian Tax Office was started in 1991 by the then-Labor government, with the automation of the system being increased in 2011. The 2015 measures reduced human oversight once discrepancies between income reported to the ATO and income reported to the DHS were identified. Previously, officers scrutinised each discrepancy on a case-by-case basis before raising an over-payment. In addition, where previously the DHS collected verifying information from employers, the new system shifted responsibility for providing information to the individuals concerned, reversing the “onus of proof”.
As the 2017 Senate committee report pointed out,
the responsibility for checking and clarifying income information has shifted from the department to current and former recipients of Centrelink payments. … the significant reduction in workload for the department by this outsourcing, has allowed for a huge increase in the number of income discrepancy investigations that the department initiates
The central problem, however, arose from the cause of the discrepancies between income reported to Centrelink and income reported to the Tax Office. Income reported to the Australian Tax Office is total income received in the financial year from 1 July to 30 June. Social Security payments are fortnightly and the level of entitlement is based on the financial and personal circumstances applying during the fortnightly reporting period for which the payment is made. People receiving a payment have their employment income assessed in the instalment period in which it is earned, derived, or received, based on whichever event occurs first, usually when people earn the money.
The payment rate to which a person is entitled can vary from fortnight to fortnight, depending on the hours of work undertaken in the reporting period and the wage rate paid. An individual could theoretically be entitled to payments in one fortnight, zero payments the next fortnight, and positive payments in the following fortnight.
Averaging will only work correctly if individuals receive exactly the same income each fortnight. This is clearly not the case for people who receive payments for part of the financial year, for example, tertiary students – like Masterton and Amato – whose study year is likely to run from March to November.
Why was Robodebt a policy fiasco?
There is an extensive literature on “policy failure”. Well-known examples range from cost overruns in constructing the Sydney Opera House and the Bay Area Rapid Transit network in San Francisco, to the poll tax in Britain in the 1980s.
What separates a “fiasco” from other forms of policy failure is that it is commonly regarded as being reasonably foreseeable – that is, the failure should or could have been avoided with foresight.
The current situation was clearly foreseeable. In July 2017, Peter Hanks gave the National Lecture at the Australian Institute of Administrative Law National Conference in which he criticised the Commonwealth for its new method of raising and recovering what “Centrelink chose to describe as ‘debts’”.
Professor Terry Carney – who served for nearly 40 years as a member of the Social Services and Child Support Division of the Administrative Appeals Tribunal and its predecessor the Social Security Appeals Tribunal – also made five judgements in 2017 that there was no debt where Centrelink calculated overpayments by applying averaged annual income to shorter periods.
In an article in 2018, Carney noted the Commonwealth had to prove that debts existed rather than requiring individuals to prove they did not. The Social Security Act requires the government to prove that a debt exists, whereas the reversal of the onus of proof effectively required people to prove that a debt did not exist. This was also pointed out by Hanks in 2017, and was the conclusion reached by the Federal Court in 2019.
The department could have appealed these judgements to the next level of the AAT where rulings would have been published, but chose not to – suggesting it was not confident of its position. The Guardian later revealed there were 76 other AAT cases where debts were set aside because the calculations used by Centrelink “could not lawfully support the existence of a debt”. In each case, the Commonwealth elected not to appeal.
Who or what was to blame?
The Robodebt fiasco involves policy failures across numerous dimensions. The most obvious – and in many ways the least important failure – is that it failed to achieve the budgetary savings that were its main objective.
A more important failure is that the Australian government decided to act unlawfully. This was compounded by the failure of the administrative procedures of government to identify and prevent these unlawful acts.
Even more seriously, hundreds of thousands of people were adversely affected. This human cost is difficult to assess and involves much more than financial losses. Just over 2,000 people who had received a Robodebt notice between July 2016 and October 2018 died during that period, although in the absence of an official coroner’s report, no causes can be attributed. However, it has been noted that from January 2017, Centrelink began tweeting the contact number for Lifeline, the national charity providing 24-hour support and suicide prevention services.
Identifying which individuals made the decisions over many years that lead to this result would seem to require a Royal Commission, as suggested by the Opposition.
While the policy decisions leading to Robodebt were made by individuals, they were made within a layered context of precedents, established processes, and social, economic and political environments that structure the choices these individuals made.
The most important aspects of this deeper political environment include the strong commitment of Australian political parties at the time to achieve a balanced Budget, the formal and informal Budgetary rules about spending and savings, the size of social security and welfare spending (around 35% of total Commonwealth spending), the highly targeted nature of the social security system, and political and popular judgements about social security recipients and their “deservingness”. An emerging aspect of this deeper background is the development of digital benefit delivery and reporting, and the growing role of automatic decision making
The confluence of these background factors can be thought of as producing a perverse “window of opportunity” – similar to what Kingdon has described as the combination of contexts producing successful public policies, but in this instance responsible for an ill-intentioned policy bypassing established processes. Kingdon points to these windows opening at the moment when a problem is recognized, a solution is available and the political conditions are favourable.
In the case of Robodebt, while the problem was presented as the need to ensure integrity in the social security system, for government the real problem was how to cut social security spending. The solution was to use a long-established system of data-matching, but one where data on historic debts was now more readily available and technological advances could be used as a screen to claim greater efficiency. The political conditions were favourable due to long-standing views by political parties and many in the public that recipients of welfare payments are undeserving and a burden on taxpayers.
It is notable that the projected savings from Robodebt were much more significant relative to the Budget deficit than the estimated “overpayments were to social security spending. In 2019, before the government conceded that Robodebt was unlawful, the Budget Papers had boasted “The 2019‑20 Budget delivers a surplus for the first time in over a decade.” The projected cash surplus in that year was wafer-thin at $5 billion. The $1.8 billion that the government has been required to pay is more than one-third of what was the projected surplus but less than 1% of annual spending on social security and welfare.
Nevertheless, while the checks on power that are built into the appeals processes in the social security system were initially brushed aside, the combination of community activism, journalistic investigation, political scrutiny and the legal aid system appear to have ultimately halted this unlawful process. In future, there is clearly a need to strengthen these accountability and review structures.