Correcting the Record? Rebutting Five Flawed Defences of the Robodebt Programme

Dr Darren O’Donovan

By Darren O’Donovan

Today sees the publication of the Senate Committee on Community Affairs’ report into the Department of Human Services’ “robodebt” programme. This system first started to wedge its way into my life at the very tail end of semester last year. Clearly tired and upset students would join the queue of questioners after my classes, to catch me privately and to ask if there was someone they could talk to about how to secure information from multiple or missing employers or what “reasonable excuse” meant. I began to recognise these students by the way they would stand back to let the other law students ask about the course content first, not wanting to be overheard in front of others. These are simple examples of the crowding out of educational opportunity by a system which was badly designed at its crucial moment of inception. An automated system tied to a time limited reverse onus procedure has a disproportionate impact on those engaged in casual employment or transitioning in and out of education. Those were key moments in my students lives – brutally put – time spent on unspooling centrelink’s calculations will not appear on their transcripts when they go for job interview.

These impacts have simply not been recorded in the closed calculus of the line item budget return for the debt recovery programme. Today, more than any technical recommendation, is a chance to mark meaning – to expressly acknowledge the unacceptable errors that were made, and to more fully reflect upon challenge of delivering on the ground fairness and balance in arriving at the calculation of overpayments. The “online compliance initiative” (the robot of public imagination) has attracted quite a care team – from a damning Ombudsman report, to ongoing and promised Privacy Commissioner audits, to something the Department of Human Services terms an “interative process” of review by Price Waterhouse Cooper, to the drafting in the CSIRO’s Data61 unit. Yet, as each of these institutions cranes over the patient in evident bafflement at the initial diagnosis, the Department reacts to widespread criticism, calls for slow downs surrounding the Initiative by using the language of “refinement”, casting actions urgently required to ameliorate chronic failures of administration as some kind of voluntary service delivery update.

Today, the stories of everyday Australians and the impenetrable documents tendered in evidence to the committee should ring out. This is my effort to square off some bullet point defences of the programme, which should not be allowed to obscure those voices.

  1. The process of data matching/averaging is a long established technique used by both sides of politics

This underplays the fact that the Online Compliance Initiative of July 2016 differed in crucial respects from that which prevailed previously. This is the timeline confirmed by the Ombudsman report:

Prior to 2015: roughly 20,000 “high risk” debts which were flagged through data matching. These would be followed up manually by members of the Department through a combination of the contact with the recipient and the Department’s own statutory powers to obtain information. The controversial averaging out of income was a last resort after this extensive, manual engagement.

In July 2015:

A new system termed “interim manual” is created. This takes 100,000 debts (so still a bit selective, not quite at the scale of robodebt) and:

  • No longer seeks information from employers in order to clarify
  • Requires the individual to supply all clarifying information within 21 days
  • If insufficient information provided, the income of the ATO would be averaged out (more on the unfairness of averaging later)

In July 2016:

Full Robodebt launched,

  • This sent letters for all debts in which a discrepancy was identified.
  • The individual was asked to supply the information clarifying their income for the past six years.
  • If information is not provided in the prescribed form, the income of the ATO would be averaged out. (more on the unfairness of averaging later)

My point is that there are key, identifiable differences which show how the system ramping up to the unprecedented storm that hit over Christmas. The system previously operated through selective tranches and was backstopped by information gathering powers. Finally, the talking point is undermined by the Department’s own action of recently calling in the CSIRO’s data team Data61, which the department has asked to produce risk profiles to better target the letters. This is stepping away from automation back to the form of tranching or risk profiling that marked the pre 2015 (and to a lesser extent interim manual) system. The Department, has however, adopted an opaque position on the use of its information gathering powers (more below).

  1. There is no error rate in the “system”. The Ombudsman found there was no error rate in the “system”. The “system” is working as planned.

Crucial thing with this point is the use of the word “system” rather than Initiative. This talking point addresses the software which produces the data match, not the end result of the Initiative. My favourite example of this is from the Department’s answer to a recent Question on notice:

“Evaluation of the pilot showed that the system calculated debts accurately.  The Ombudsman also noted in his recent report that the Online Compliance system calculates debts accurately based on the information available.”

The crucial difference between those two sentences just smacks you in the face. In the media (particularly in the reactions to the Ombudsman’s report) there is a tendency for crucial clause in bold to be missed, dropped or otherwise undervalued. But as I’ve just explained – robodebt is a combination of a reverse onus approach to information and the automated issuance of discrepancy letters on scale beyond the previous regime targeted at “high risk” debts.  The Ombudsman report found that taken as a whole, the initiative, was not in accord with administrative law principles of reasonableness and fairness. It is misleading to simply plunder the report’s discrete finding on the robot which is a prisoner to the Department’s poor dataset choices. The first priority item under the Ombudsman’s terms of reference was to check what the robot was doing – part of the system – what had triggered all this outrage. As you work through the Ombudsman report you slowly see that the key issue is that the Department had locked its lonely robot away and fed it a starvation diet.

While the term “robodebt” has thus been very useful to communicate the automatic, mass nature of the letters, its downside is that the department has often segmented the analysis of the system. The Department’s public positions often focus discretely upon the “robot” data match, inferring that the “opponents said the robot was at fault”, but that the Ombudsman found robot calculates things accurately. This has led to frankly bizarre exchanges before the Senate inquiry about “error rates”, and blanket assertions that there is no “error rate”.

The key question is whether when the debt is issued it is done so with sufficient certainty, having taken all relevant considerations into account and having accorded the person adequate procedural fairness. As the journalist Justin Warren’s running spreadsheet underlines, there is an established high rate of the initiated debts being changed, recalculated and withdrawn. Exasperating the Department’s attempts to frame these changes as not being “errors” but rather the individual doing what they were told to do. At times, it feels like there is no real limit point to this reasoning, that in effect it is near impossible for the Department to issue a debt in error. Even the high rate of successful appeals to the AAT are, in the Department’s eyes really only due to “new information” “emerging”. There is a danger that the Department is becoming analogous to a teacher who rather than teaching the class, merely grades papers at a fail and waits passively for the redrafts. This may be a cost-efficient approach, but we’d all better have much more realistic and detailed conversation about it as a society.

  1. We are seeking clarification, these are after all only clarification letters.

Data-matching with ATO information was always going to be imprecise, underinclusive of key information and not adapted to the modern realities of casualization and seasonal work. The robot’s function inbuilt income averaging, traditionally a last resort for a decision-maker who had interacted with the individual, to hundreds of thousands of debt estimates. The data match was inevitably primed to overcalculate the debt, and then teamed with the idea that the onus is on the individual to provide all other information. The Department’s own internal operational documentation (‘Acceptable documents for verifying income when investigating debts’) strongly underscores the potential unfairness of averaging in the era of casual or flexible work:

If employment is for a part of the year only, averaging over 12 months will not result in a correct result if the customer should have received a full rate at other times of the year.

If income varied greatly during the year, the result may be incorrect.

While data matching is pretty straightforward, the Department has accepted that explaining why a data match may be underinclusive is very complex, and at first instance, it’s very difficult to explain the calculation (rather than simply asking for particular documents). As a Deputy Secretary explained at one hearing:

…we are dealing with quite a complex social security act—different payments, different rules—and then the intersection with the tax act about income. So we are trying to explain quite a complicated scenario. Then, if you overlay that with the individuals themselves, some of them have quite complicated lives, or different things happening in their lives, which can also add to the complexity. It is not something that is easily understood. We try and make it as easy as we can, but there is inherently a lot of complexity for some people.”

Abstract discussion of the non-existence “error rates” cannot obscure the fact that the Ombudsman’s report was frankly damning of the “clarification” phase of OCI. Perhaps the clanger that should live longest in infamy is the failure to include the specific phone number for debt collection in the original letter for the first six months of the initiative. This likely loaded up the already heaving generalist phone lines, contributing to the headlines regarding 42 million unanswered calls last month. Just this week, the Department, in an answer to question on notice, the Department admitted that under their own models, they only ever expected 65% of people to use the online portal – and yet they did not include the phone number. Further flaws found by the Ombudsman:

  • It recommended that DHS should further support customers to gather employment income evidence to maximise the accuracy of possible debts.
  • It highlighted the especially onerous nature of requiring a person to retain or access employment and payroll records for a period of six years, without forewarning. The trigger on the exercise of information gathering should be informed by the fact ATO only requires individuals with simplified tax affairs to retain records for two years.
  • It found there was a failure to advise clients on what they could do if they had problems obtaining evidence.
  • The Office’s finding that “it is critical for DHS to give some customers additional support and assistance to obtain this evidence when they have made genuine and reasonable attempts and other available information is not sufficient.”
  • The initial failure to expand the modes of communication to registered post or other modes which were clearly more adapted to confirm receipt and understanding of the discrepancy letter.
  • Rigid reference to internet after the individual had clearly signalled possible vulnerability/inability to use this.

In responding to the Senate Inquiry it is important to foreground the track record of the Department in implementing Ombudsman’s reports, (especially the infamous phoneline run around of 2014). While the Department asserts the acceptance of the recommendations, it’s accompanying narrative, to my eye, read down many of the recommendations and was eager to cast them as voluntary refinements rather than critical, legally required steps for a system in crisis.

  1. These letters are not sent out to vulnerable people

Simply pointing at the robot, and throwing the onus on the individual to unpick its estimate, was a decision not supported by empirical research into Australians’ legal and functional literacy. The Department cannot say that no vulnerable individuals received letters – the reality was that no individuals flagged as vulnerable in Centrelink’s system received a letter. The Ombudsman’s Report supports the view that no one in the Department stopped to specifically consider in what circumstances an individual would be too vulnerable to embark on the intense process of contesting a debt. The Ombudsman report found the vulnerability flagging system of the Department would not capture individuals who were not in regular contact with Centrelink (the main population for the letters), it would not capture individuals who were in ongoing situations of danger, those with mental illness aggravated by receipt of the letters. For these individuals, the practical result in the first months was not alone the locking of the debt by averaging, but also a further 10% financial penalty being applied for the debt recovery process being set in train. The unfairness of that penalisation was underlined when it emerged that someone in the Department took the extraordinary action of recalling over 56,000 debts in February. This was done to advise individuals of their review rights, but the fact that the Department still stands over the debt recovery penalties, shows a fundamental refusal to accept the unfairness the system as it operated from July to January.

The crucial question of when the Department will help people by using its statutory powers to obtain information, received entirely opaque treatment in the Department’s responses:

“DHS will use its powers to obtain information on a case by case where other avenues have been exhausted”

I have seen no on the record explanation of what “case by case” means. The Department has the ability (and prior to 2015 this was used) to issue a statutory notice to gain information from employers or any holding entity holding information relevant to a debt. Over the course of the past six months, we have instead seen individuals across Australia scramble to get information from multiple employers who went bust back in 2011/2012 or educational records from VET sector providers now gone into administration. A great example is the individual who, in order to get the debt recalculated, has to prove that they did not teach as a TAFE over Christmas 2012. There is a vast universe of logically probative evidence sources that would support the stripping out of that fortnight from the averaging of income. These should be evaluated before the debt is raised. Furthermore, while bank statements can get you a net pay figure, reverse calculating the gross pay figure Centrelink requires to calculate payments is of absolutely eye-watering complexity (think tax allowances, deductions etc). People are encountering significant difficulties in using bank statements and group certificates to reconstruct their gross income. The idea that current arrangements can be seen as a good faith, co-operative mechanism of clarification is not supported by the practical reality of this scheme.

  1. This is about integrity: seeing that people get what they are entitled to, no more no less”

  1. Right out the gate this talking point doesn’t really survive the finding of the Ombudsman report that:

“We asked DHS whether it had done modelling on how many debts were likely to be over-calculated as opposed to under-calculated. DHS advised no such modelling was done”

  1. The “no more, no less” principle is functionally obliterated by information recently supplied to the senate committee by the ATO that no individual who has started repaying a robodebt has applied for a reassessment of their tax liability despite their being entitled to it. This points to a massive access to justice issue – and the need for a whole of government approach to settle the question of income calculation. The Department of Human Services has run a programme that does not reverse data match to ensure consistent calculation across government services. The Australian Taxation Office may now face serious budget implications as it may have to readminister hundreds of thousands of taxation assessments. This policy is totally out of alignment with the ATO’s approach to time limits in the pursuit of tax liabilities (outside of deliberate fraud and evasion). The process of getting any tax back is currently very unwieldy with limited public information available.
  1. The Administrative Appeals Tribunal which has seen a large jump in the level of appeals in Centrelink debt decisions, is now required to “absorb” this spike under its current budgetary allocation. Far frin being a measured, properly integrated government programme aimed at efficiency, the costs of robodebt are being externalised onto other governmental actors.
  1. Canberra’s primary logic is the internal line item budget. It is only within that closed calculus that this programme that the identified hundred of millions of savings can be asserted. The senate inquiry showed just how frontline social services and community legal centres, the key to so many society’s crisis situations were hit by a wave of robodebt work. The opportunity cost of that work is extraordinary. And this is before we get to the lost productivity of individuals trying to unravel Centrelink’s reasoning.
  1. Finally, the testimony of Greg Barns, who represented Australian Lawyers Alliance at the senate hearing, underlined that the OCI initiative may actually place the system we have for investigating serious welfare fraud under strain. His testimony described the quality of files being sent for prosecution in serious fraud cases is poor, with serious consequences for trial and sentencing.

I believe that the experiences described by Australians to the Senate inquiry cannot be regarded inevitable uneven remainders of a fair process of public administration. To describe the remedial actions now required as minor refinements or continuous service improvements represents a profound undervaluing of the centrality of administrative law principles and scaffolds to the daily practice of government and service delivery. The Department needs to be held accountable for its failure to ensure a practical, realistic translation of the right to procedural fairness for key, identifiable categories of recipients, especially the most vulnerable.

Further information to clarify any of my points further is available in my submission to the inquiry.



La Trobe