The final report of the Royal Commission into the Robodebt Scheme comments
Robodebt was a crude and cruel mechanism, neither fair nor legal, and it made many people feel like criminals. In essence, people were traumatised on the off-chance they might owe money. It was a costly failure of public administration, in both human and economic terms.
The report features damning comments on individuals and agencies, including the Office of the Australian Information Commissioner and the Commonwealth Ombudsman's Office.
The report states
There are different mindsets one can adopt in relation to social welfare policy. One is to recognise that many citizens will at different times in their lives need income support - on a temporary basis for some as they study or look for work; longer-term for others, for reasons of age, disadvantage or disability - and to provide that support willingly, adequately and with respect. An alternative approach is to regard those in receipt of social security benefits as a drag on the national economy, an entry on the debit side of the Budget to be reduced by any means available: by casting recipients as a burden on the taxpayer, by making onerous requirements of those who are claiming or have claimed benefit, by minimising the availability of assistance from departmental staff, by clawing back benefits whether justly or not, and by generally making the condition of the social security recipient unpleasant and undesirable. The Robodebt scheme exemplifies the latter. ...
In September 2013, the Liberal-National Coalition, led by the Hon Tony Abbott MP, won government. Interviewed by the Australian Financial Review in December 2012, Mr Abbott had made some expectations clear, in what the writer described as a “blunt warning to public servants:” Because what normally happens is that a government comes in, they’ve got all these policies and they rely for the implementation on the public service ... There’s nothing wrong with that as such, but what I’d like to be able to say to the public service is, ‘Look, this is how we think it needs to be done’. Rather than relying on them to tell us, I’d like to be in a position to tell them on day one.
The executive government depends on the public service for its administration, but the more authoritatively the government of the day can speak to the public service, the more likely it is that the public service will administer your policy in a way which recognises its spirit and its letter.
Budget control and debt reduction had been second in the Coalition’s list of policy priorities in its election manifesto. Consistent with that policy, in July 2014 the Hon Kevin Andrews MP, the Minister responsible for the Department of Social Services (DSS), proposed the setting up of an interdepartmental committee to develop a whole-of-government strategy for recovery of debt owed by members of the public to the Australian Government. The terms of reference included examining data matching, using online and self-servicing options, using external debt collection agencies and applying a standardised interest charge to debts. And in relation to welfare services, in January 2015 the newly-appointed Minister for Social Services, Mr Morrison described himself in an interview as planning to be a “strong welfare cop on the beat;” because Australians were “not going to cop people who are going to rort [the social security] system.”
It was in this climate that the essential features of the Robodebt scheme were conceived by employees of the Department of Human Services (DHS), were put by way of an Executive Minute in February 2015 to the Minister for Human Services, Senator the Hon Marise Payne, and to Mr Morrison as Minister for Social Services. Approved by the latter, they made their way in the form of a New Policy Proposal (NPP) through Cabinet with remarkable speed. In May 2015, as part of its 201516 Budget, the government adopted a measure named Strengthening the Integrity of Welfare Payments. Described as a package for “enhancing ... fraud prevention and debt recovery and improving assessment processes” in relation to the payment of social security benefits, it was expected to save $1.7 billion over five years. Most of those savings were to come from the Employment Income Matching measure, the initiative which began Robodebt, which was proposed to recover overpayments resulting from incorrect declarations of income. Another measure in the package, titled “Taskforce Integrity”, involved the secondment of Australian Federal Police officers and was designed to crack down on welfare fraud. The two were often, and not coincidentally, mentioned in the same breath.
In summary, the Employment Income Matching measure entailed a process of data-matching and debt raising to be applied to some 866,857 instances of possible overpayment for the financial years 201011, 201112 and 201213 identified through a comparison of Australian Taxation Office (“ATO”) and DHS data. DHS would obtain information from the ATO as to what a benefit recipient’s employers had returned as the income earned by the recipient in the relevant financial year (“PAYG data”), compare it through an automated system with what had been declared to DHS by the recipient, and in the event of discrepancy would require the recipient to go online to explain it.
If the recipient did not respond or provide details which explained the discrepancy, or agreed with the PAYG data, the amount declared by the employer would be averaged across a period of time, which was often whatever employment period the employer had indicated. The system applied some rules which attempted to account for the varied circumstances of income support recipients; however, the basic premise was that the amount declared by the employer would be divided up evenly, and allocated into the number of fortnights included in the period being reviewed. (That process was variously referred to income “averaging”, “smoothing” or “apportioning”.)
Where, in any given fortnight, the averaged fortnightly amount exceeded the income the recipient was entitled to receive before reduction of benefit, it would be taken that there had been an overpayment, a debt would be raised accordingly and steps would be taken to recover it. Where the recipient was still on benefit, deductions would be made from the income support currently being paid. Where the recipient was no longer on benefit, they would be required to enter a repayment arrangement. Debt collectors would be involved if they did not respond, and they were liable to have any income tax refund they were entitled to receive garnished. Robodebt began with some pilots in 2015, was rolled out with its online platform in September 2016 and continued in various iterations for four years.
In a broad sense, certain elements of the Robodebt scheme were not new. DHS had for some years undertaken data-matching with the ATO to identify discrepancies between employer-reported and recipient-declared income. In the past, such discrepancies had served as a trigger for a manual review of those files exhibiting the strongest likelihood of overpayment and the highest levels of discrepancy (about 20,000 such files a year). Those reviews involved a DHS compliance officer checking the file for relevant information, contacting the recipient to see if any discrepancy could be readily reconciled and, if not, requiring their employers to provide more specific payment information. The raising of a debt and its recovery might then follow.
There were five significant differences under Robodebt.
The first and major difference was that the PAYG data was to be regarded as the primary source of earned income information, which could be acted on to raise debts although unconfirmed by the employer or the recipient.
Secondly, where in the past compliance officers had engaged with recipients, examined the file and used powers under the Act to seek information from employers in order to determine whether a debt existed, now recipients were given a gross income figure with a period which might or might not accurately reflect the period worked, with the onus placed on them to provide details to contradict it, or have a debt and, prospectively, a 10 per cent penalty automatically raised against them.
Thirdly, where averaging had previously been used to arrive at a fortnightly income figure in limited circumstances and often with the agreement of the recipient that it gave a figure which reflected his or her actual income, the PAYG data averaged on a fortnightly basis for the declared period was now applied automatically where alternative information was not provided and accepted. That was despite the fact that entitlement to income support was to be determined under the Social Security Act 1991 by reference to the actual fortnightly income of the recipient, and it could not safely be assumed that recipients earned precisely the same amounts each fortnight they worked or that they worked every fortnight during the period nominated by an employer (which did not need to be more precisely stated than the relevant financial year). In basing entitlement on actual fortnightly income, the Social Security Act reflected the policy aim of social welfare payments: to support people in their periods of real need so that a recipient who earned little or nothing in a given fortnight would be assisted then.
Fourthly, critical to achieving the predicted savings, where previously compliance officers had been involved, it was now anticipated that in most instances the entire process would be conducted online, the aim being to reduce (and preferably obviate) human involvement at the DHS end.
The fifth and final difference: where data-matching and consequent reviews had in the past been conducted on an annual basis, now recovery efforts would be directed over several years, going back some five years from the implementation of the program. (The measure was extended to include the 2013-14 and 2014-15 financial years in the 2015 MYEFO, and the 2015-16, 2016-17 and 2017-18 years in the 2016-17 MYEFO.) This seems to have been based on a notion that because discrepancies in the hundreds of thousands had been identified in previous financial years there must also be debts in the hundreds of thousands of dollars in those years awaiting recovery; ignoring the facts, firstly, that a discrepancy did not necessarily indicate an overpayment and secondly, that the 20,000 or so files involving discrepancy which had already been reviewed for each of those years had probably already yielded the bulk, in monetary terms, of the overpayments in those years.
In late 2014, in response to DHS’s initial canvassing of the Robodebt concept, DSS had obtained an opinion from its employed “in-house” lawyers. Their advice had emphasised the requirement in the statutory benefit entitlement rate calculators to consider actual fortnightly earning or receipt of income, expressing concern that averaging might, therefore, not be consistent with the legislative framework. A policy advice given at the same time similarly pointed out that calculation by averaging did not accord with the legislation and a debt amount calculated in that way might be wrong.
In February the following year, DHS officers provided the Ministers for Social Services and Human Services with the Executive Minute containing a number of proposals including Employment Income Matching (Robodebt). It pointed out that (consistently with its 2014 advice) DSS had advised policy change might be, and legislative change would be, needed to implement the Employment Income Matching initiative. On 20 February 2015, Mr Morrison signed the Minute, indicating his agreement that initiatives including Employment Income Matching be developed as a package of NPPs and that DHS work with DSS to advance consideration of the necessary policy and legislative change.
By 3 March 2015 an NPP reflecting the Employment Income Matching initiative had been prepared for inclusion in an exposure draft of a Social Security Portfolio Budget Submission. It contained no reference to legal risks and said that legislation was not required. The Scheme was approved by Cabinet and proceeded without the legislative change to support averaging which DSS had said was needed. That awkward question was avoided in the NPP by the simple expedient of not mentioning averaging and saying instead, falsely, that the new approach would not change how income was assessed or payments calculated.
Given that it was still proposed to undertake 866,857 compliance reviews (or “interventions”) for the 201013 financial years producing gross savings of $1.1 billion (as compared with an estimated $1.2 billion in the Executive Minute) and given the speed with which the NPP was developed and approved for presentation to Cabinet, it is not credible that anyone closely involved with the measure could have believed that there had been some fundamental change to the proposal so that it no longer entailed averaging or for some other reason ceased to require legislative change. It is worth considering for a moment what legislative change to enable the reliance on averaging inherent in the Robodebt scheme would have entailed: a retrospective change to the statutory basis on which social security recipients had been entitled to receive, and had received, income support going back years. It would certainly have encountered parliamentary and public opposition.
The use of averaging was by no means the only in the Robodebt scheme. No consideration seems to have been given to the legal basis on which DHS could ask recipients to provide information in response to it. The letters sent to recipients during the various iterations of the program used language indicating that a response was required, not just requested (“You need to tell us ...”) but they were not said to be notices from the secretary of DHS requiring information under any of the provisions of the Social Security (Administration) Act which enabled information to be obtained in that way. Similarly, there is no evidence that any thought was given to the grounds on which a penalty of 10 per cent (euphemistically called a “recovery fee”) was added to the debts raised against recipients. In subsequent internal legal advice, there seems to have been an assumption that if averaging resulted in a debt, it could be assumed that the recipient must have failed to inform DHS of a change in their income. That was an extraordinary assumption. The notion that averaging without more could prove a debt was unsound in the first place, but to charge a penalty required a conclusion that the recipient had committed a breach of the Social Security (Administration) Act by failing to report a change in income, and required correspondingly compelling evidence to justify it. That seems not to have crossed anyone’s mind.
Those were respects in which the Scheme was devised without regard to the social security law. More fundamentally, the way averaging was used in the Scheme was essentially unfair, treating many people as though they had received income at a time when they had not, and did not need support when they did, with the further fiction that they now owed something back to the government. It subverted the rationale on which income support was provided in the first place: as a safety net to ensure that people received help when they most needed it.
There were other fundamental unfairnesses in the program. No regard was had to the sheer unreasonableness of placing the onus on recipients to attempt to establish what their earnings were for periods going back as long as five years when they had been given no reason to expect anything of the kind at the time they declared their income and received their benefits. (DHS’s own website, at least at the start of the Robodebt scheme, contained the information that it was necessary to retain payslips only for a period of six months.) Some recipients were unable to produce payslips because employers had gone out of business or were unhelpful. Others were reluctant to approach employers for a variety of reasons, including poor relations with them, an unwillingness to trouble them, or embarrassment at having to disclose receipt of welfare benefits.
The system was set up with the intention of forcing recipients to respond online to the PAYG data and to minimise contact with DHS officers, in the interests of economy; this was vital to the anticipated savings. (Indeed, having had the benefit of “behavioural insights,” DHS employees setting up the system made a conscious decision not to include any telephone number for the Customer Compliance division in the letters sent to recipients, so as to force them to respond online, while compliance officers were told to direct recipients online.) No outside parties with an interest in welfare were consulted in order to understand how the Scheme might actually affect people. There appears to have been an obliviousness to, or worse a callous disregard, of the fact that many welfare recipients had neither the means nor the ability to negotiate an online system. The effect on a largely disadvantaged, vulnerable population of suddenly making demands on them for payment of debts, often in the thousands of dollars, seems not to have been the subject of any behavioural insight at all.
Some of the difficulties in the system in its initial online iteration (the Online Compliance Intervention (OCI)) were the result of its being put into operation in haste, rather than conscious decision-making. A pilot and a manual version of the program were conducted in 2015, using manual rather than automated intervention but relying on the PAYG data to raise a debt in the absence of a response. The online component of the Scheme was the subject of a limited release in July 2016, but it was fully rolled out in September of that year although no proper evaluation of the pilot or manual program had taken place and there were numbers of unresolved problems. The system sometimes deleted details of previously declared income; there were failures in employer name matches which led to wrong duplication of earnings; particular allowances and payments were wrongly treated. No user testing had been done on the screens presented to recipients when they responded to the DHS initial letter; that was supposed to be done in 2017. Recipients had difficulty uploading data and from time to time the platform went down altogether. It was not just recipients who struggled with the system; so did DHS compliance officers.
Other problems were inevitable. Former recipients were contacted through the addresses held for them on DHS records – a postal address or a myGov account – although, since the debts went back some years, they might well have moved away or ceased to monitor their myGov account for any Centrelink notices. Their inevitable failure to respond resulted in income averaging being applied and debts raised against them. The 2015 pilots showed that only about 40 per cent of customers were making contact. On that evidence it could reasonably be expected that a majority of recipients would, when the automated system began, be the subject of automated debt raising using averaging because they did not receive the notifications, did not understand what they needed to do, or thought their dealings with Centrelink were long past. Others did log on and, as the initial letter instructed, confirmed the information presented – the employer’s name and the amount of earnings in the period nominated by the employer – on the basis that it did represent their income for the financial year in question. They were given no warning that their answer would then be used to raise a debt against them by way of averaging over the employer- nominated period, often the entire financial year, even where they had worked for shorter periods. (DHS’s own figures showed that as at 27 January 2017, 76 per cent of those who were subject to the OCI had debts raised based on averaging of their earnings: 99,404 recipients). And no-one could understand how the debts were calculated, because the debt notices gave no explanation.
The disastrous effects of Robodebt became apparent soon after it moved, in September 2016, from the last part of the limited release, involving around 1000 recipients, to sending out 20,000 notifications per week. In December 2016 and January 2017 the media, traditional and social, were saturated with articles about people who had had demonstrably wrong debts raised against them, and in many instances heard of it first when contacted by debt collectors. The human impacts of Robodebt were being reported: families struggling to make ends meet receiving a debt notice at Christmas, young people being driven to despair by demands for payment, and, horribly, an account of a young man’s suicide. The Australian Council of Social Services, the peak body for community services supporting recipients, wrote to the Minister for Human Services in December 2016, pointing out the inaccuracies which were being produced by averaging instead of applying actual fortnightly income figures, the unfairness of charging a penalty where it was not established that a recipient had even been contacted, the difficulty for people in recovering information from employment years past, the technical difficulties with the online system, the lack of assistance from Centrelink officers and the commencing of debt collection often without warning to the recipient.
The beginning of 2017 was the point at which Robodebt’s unfairness, probable illegality and cruelty became apparent. It should then have been abandoned or revised drastically, and an enormous amount of hardship and misery (as well as the expense the government was so anxious to minimise) would have been averted. Instead 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. The government was, the DHS and DSS ministers maintained, acting righteously to recoup taxpayers’ money from the undeserving.
DSS obtained cover in the form of what was called a “legal” advice supporting averaging from one of its in- house lawyers. That advice expressed the view that it was open to the DHS secretary “as a last resort” to act on averaged income to raise and recover a debt where a recipient did not provide income information after being given an opportunity to do so. No legislative provision or case law was cited to support that proposition. That advice – but not the 2014 advice pointing out the inconsistency of averaging with the legislative framework – was provided to the Commonwealth Ombudsman, who was conducting an own motion investigation into the Scheme, including a consideration of its legality.
When the Ombudsman pressed for any advice DSS had given about legislative change needed for the use of ATO data prior to the February 2015 Executive Minute, senior DSS officers provided the 2014 advice and offered this justification of its obvious inconsistency with the 2017 advice. After the 2014 advice was given, DHS had adjusted the process, giving recipients the opportunity to correct the PAYG data, so as to assuage DSS’s concerns and satisfy it that legislation would not be needed. This explanation suffered from two deficiencies: it was untrue – the opportunity to correct information had always been part of the proposal – and it made no sense. DSS reinforced its position with a letter from its secretary to the Ombudsman asserting DSS’s satisfaction that the system operated “in line with legislative requirements” and that DHS had made no changes to the way it assessed PAYG employment income.
Uncompelling though DSS’s 2017 advice and explanation were, the Ombudsman decided against raising questions in his report about whether averaging under the scheme had legislative authority. With a good deal of input from DHS employees, he reported in April 2017 that, while there were numerous technical problems with the operation of the OCI, after examination of the underpinning business rules, he was satisfied that it raised accurate debts based on the available information. Those business rules, the Ombudsman concluded, “accurately capture[d] the legislative and policy requirements.” Thereafter, ministers and departmental representatives relied on the Ombudsman’s report as proof of the legality and appropriateness of Robodebt against all comers: the media, opposition politicians, social welfare bodies, critics in academia and a Senate Committee which made adverse findings.
(In 2018, the Ombudsman again raised concerns about the legality of averaging but was fobbed off with explanations that it involved the best available evidence and was only used where the customer had not taken the opportunity to provide information. DHS dissuaded the Ombudsman from mentioning the legality issue at all in his 2019 report into implementation of the recommendations in the 2017 report.)
Meanwhile, the Scheme trundled on, with the government engaging PricewaterhouseCoopers to assist with some of its clumsier components (although never taking receipt of a critical report prepared by the consultancy). It had to accept that the Scheme was not functional in many respects. One was that the online component was an abject failure, with the result that large numbers of employees had to be drafted on short-term contracts or by way of labour hire into the DHS to cope with enquiries. And it became apparent, partly as a result of that factor, but also because of the overestimation in the first place of the numbers of debts and their average amount, that the touted savings would never be reached.
The cover-up continued. Social Security recipients against whom a debt was raised could seek review from the first level, or Tier 1, of the Administrative Appeals Tribunal (the AAT). To do so required some understanding of what had actually happened, which was virtually impossible to ascertain on the DHS documents, as well as the confidence and capacity to seek review. Still, a number of decisions was made by AAT members setting aside debts raised under Robodebt in its various incarnations for the reason that averaging, of itself, could not provide evidence of actual income or, it followed, overpayment and debt. Section 8(f) of the Social Security (Administration) Act requires the DSS secretary to apply government policy with due regard to relevant decisions of the AAT. If the secretary disagreed with them, Tier 1 decisions could be appealed to Tier 2, a higher level of the AAT. Arrangements between DHS and DSS required that DHS refer AAT decisions to DSS where there was a significant error of law, a significant issue of policy or administrative practice, or the matter had attracted or was likely to attract media or parliamentary attention. And DHS and DSS had adopted joint litigation principles, which required that the decision whether to appeal any decision would be made with regard to whether there was need for clarification of a significant point of law or stated Government policy.
The AAT decisions setting aside Robodebts on the ground that evidence of averaging was not capable, without more, of proving a debt met those criteria. Because DHS disagreed with them, it ought to have referred them to DSS for consideration of appeals and it certainly should not have continued to use averaging in disregard of what was said in them. However, DHS took the course instead of taking whatever steps were directed by the AAT to rectify the individual cases by obtaining other evidence, but otherwise ignoring the decisions. The fact that Tier 1 AAT decisions were not published made it easier for it to do so.
DSS seems to have taken little interest in the AAT cases until mid-2018, when DHS referred an AAT decision that income averaging was unlawful to it. The AAT member had placed reliance on an article by Professor Carney, a former AAT member, raising a number of legally-based criticisms of averaging. The combination of the case and article prompted a decision to obtain an advice from solicitors, Clayton Utz, as to the lawfulness of using averaging to determine income. The draft advice which the solicitors provided said that it was not permissible to determine a recipient’s entitlement to benefit by averaging employer-reported income from the ATO. It was received in August 2018 and should have prompted, if not the immediate suspension of Robodebt averaging, at the least the immediate obtaining of further advice from the Solicitor-General. However, it was never put into final form or acted on.
It took two applications for judicial review made to the Federal Court in 2019, Masterton and Amato, to finally change the government’s position on Robodebt averaging, but that did not come quickly. In relation to the first, Masterton, lawyers with the Australian Government Solicitor’s office advised DHS in March 2019 that, given that the applicant had not been receiving a consistent fortnightly income, averaging could not provide an accurate assessment of her income or establish the existence of a debt, and there was no statutory basis for deeming that it could. DHS then had two options: it could run its averaging argument and get a Federal Court ruling as to its legality or it could recalculate on other evidence to arrive at a nil debt figure in the hope of ending the proceeding. Digging in, DHS chose the second course but, at the Australian Government Solicitor’s suggestion, took steps to obtain further advice from the Solicitor-General. In September 2019, the Solicitor-General advised that averaged PAYG data could not support a conclusion as to the amount or existence of a debt in the absence of further evidence that the recipient earned a consistent fortnightly income over periods precisely aligning with the periods declared by the employer to the ATO. That advice led to the settlement of the second application for review, Amato, with an accompanying declaration by the Federal Court that averaged PAYG income information was not capable of satisfying a decision-maker of the existence of a debt. Shortly before the settlement of Amato was formalised, on 19 November 2019 the Minister for Government Services announced that debts would no longer be raised solely on the basis of PAYG data. On the same day, recipients present and past, against whom debts had been raised using the averaging process, commenced a class action against the Australian Government. (Soon after, DHS, which had been renamed Services Australia, ceased to exist, being recreated as an agency within the Social Services portfolio.) In May 2020, the Solicitor-General advised that the government was bound to fail in defending the claim for unjust enrichment made in the class action. That was followed by the government’s announcement that it would refund debts raised wholly or partly on the basis of averaging if they had been repaid, and would reduce them to zero (“zero” them) if they had not. This involved reimbursement of $746 million to some 381,000 affected individuals and writing off debts amounting in total to $1.751 billion. In November 2020, on the day it was set for trial, the government settled the class action.
Declaring income earned was not necessarily straightforward for income support recipients because their earnings might be irregular and the periods over which they earned would not necessarily match the Centrelink fortnight on which income was calculated. Inaccuracies could be hard to avoid. But DHS did nothing by way of investigation of discrepancies before demanding explanations from recipients. For people who were able to prove they did not owe a debt, it was a stressful and time-consuming process. Undoubtedly some people paid amounts they did not owe because they were not in a position, practically or psychologically, to demonstrate otherwise.
For other people who might have owed what would, certainly to them, have been significant amounts, the process was unreasonable: suddenly and unexpectedly being confronted with demands for information and payment in respect of benefits which might have been received and spent long ago. People who did owe some amount were unable to get any clear information as to what they owed and why. Of those people who were overpaid, it is questionable how many of them owed debts at a level which justified interrupting their lives years later to demand repayment and it is unknown how many of the debts recovered during the life of the Robodebt scheme actually proved, once payslips were provided, to have been of too small an amount to meet the cost of recovery.
