Go to top of page

Robo-debt And Denial: The Protocols Of Centrelink

21 June 2017
Ben Eltham
New Matilda

This is the third feature in Ben Eltham’s 2017 investigation into Centrelink’s robo-debt program. The first article in the series was published in January, and the second article in March.


Victoria Legal Aid’s Dan Nicholson has a good story about the Department of Human Services and its sprawling and opaque bureaucracy. “I guess I would characterise it as an extraordinary lack of transparency,” he told New Matilda.

You could also call it the “Showdown at the Louisa Lawson Building”.

Lawyers at the Victorian legal aid agency wanted to get a copy of Centrelink’s robo-debt algorithm.


As a source inside Centrelink told me in December:

“… the earnings might be spot on and recorded correctly, but the automated system is not going to be able to reconcile all of those instances with a data match. [It] is inevitably going to end in a letter, even though the calculations during the time on benefits were likely correct.”

Robo-debt started small, but it ramped up very quickly. Although it officially began in July last year, the robo-debt program didn’t get going in earnest until September. By December, 20,000 robo-notices were being sent out each week. The government expects to send out nearly 1.7 million notices in the coming three years.

As the notices started flowing, recipients started getting issued with crippling debts. Their benefits were also being cut off. Desperate people besieged welfare agencies around the nation.

People issued with debts found it almost impossible to understand or argue their case. Unable to navigate the usability maze of the federal government’s MyGov website, or get through on Centrelink’s dysfunctional phone lines (there were 42 million unanswered calls in 10 months), clients started flooding into welfare providers looking for help.

By January, the program was in full swing. Debt collectors were actively pursuing at least 6,600 welfare recipients who were later found to have no debts outstanding. Local MPs such as Andrew Wilkie started to hear about it too, as panicked constituents contacted their federal members. Such was the impact in Tasmania, the state’s welfare agencies created a disaster relief fund to deal with it, essentially declaring a public emergency.

By February, an Ombudsman’s investigation had been announced and a Senate Inquiry hastily instituted. By April, Victoria Legal Aid was representing one of the welfare recipients caught in Centrelink’s vast new dragnet.


Victoria Legal Aid requested access to the so-called ‘Program Protocol’ – the technical document, issued by the department of Human Services, that describes the operation of the robo-debt algorithm.

“We wrote and the Minister declined to provide us any of the documents,” Nicholson told us in a phone interview. “We then noticed that the Protocol, according to the Government Gazette, should have been available in Canberra.”

Lawyers and activists wanted to see the Protocol because they couldn’t understand how Centrelink was getting it so wrong. They hoped the Protocol would uncover crucial information detailing exactly what the Department of Human Services was doing with its data matching, and what sorts of data Centrelink was trying to match up. After all, Victoria Legal Aid was representing clients appealing Centrelink decisions issued through the robo-debt program.

But looking for the algorithm proved much harder than anyone anticipated.

On the face of things, it should have been a simple matter of asking the Department of Human Services. The DHS had even issued a public notice in the Australian Government Gazette in August last year, explaining that it was abiding by guidelines from the Office of the Australian Information Commissioner.

The robo-debt controversy has placed Australia’s Information Commissioner, Timothy Pilgrim, in a fascinating position. The Information Commissioner exists as an independent statutory agency within the Attorney-General’s Department. The agency is a regulator with legal powers to investigate breaches of the Privacy Act. It also enjoys special legal responsibilities in regards to government data matching. Government agencies like Centrelink are required to comply with a set of data matching guidelines, one of which is that it should publish a written protocol of the data matching program.

Given that, you would have thought that the protocols would be available on a DHS web page. But they were not online.

The snippet in the Government Gazette had noted that a copy of the protocol was available on request from the “Case Selection Section” on Level 2 of the Louisa Lawson Building, the Department’s $225 million neo-postmodernist office building in Cowlishaw Street, Greenway. So that’s where the legal aid laywers went next.

But when Victoria Legal Aid rocked up to the foyer of the Louisa Lawson Building, they were politely told they could not have the document. They weren’t allowed up to Level 2.

“Level 2 of the Louisa Lawson Building is not accessible to the public, nor is it possible to request access to Level 2 from the Ground Floor of the Building,” they wrote in their submission to the Senate Inquiry on robo-debt. The luckless lawyers were told that “only the Director of the Section could authorise access to the document, and the Director was not available”.

Someone would be in contact, they were told. But no-one ever did contact them back. “There was no subsequent communication from any staff of the Department regarding access to the Program Protocol,” they wrote in their submission.


I waited a fortnight, then issued a Freedom of Information request. By mid-May, nothing had been forthcoming, despite the Senate Inquiry beginning and holding public hearings around the country.

Then, on May 17, I received an official “communication” from the DHS freedom of information team. “Although the department is deemed to have refused your request by operation of the FOI Act, we are nevertheless working to provide you with a decision as soon as possible,” they wrote.

It would seem as though my FOI request had been denied. “Please provide your BSB and Account Number so that the department can issue a refund of the charges paid by you,” the email concluded.

But the protocol had already been released – the day before, in fact. On May 16, Information Commissioner Timothy Pilgrim had already disclosed the protocol in a letter responding to Nick Xenophon Team Senator Skye Kokoschke-Moore. Kokoschke-Moore had written to Pilgrim in April, after hearing about the showdown in the foyer of the Louisa Lawson Building.

Sure enough, on May 22 the DHS emailed me back with a copy of the mysterious protocol.

The irony of the wild goose chase? It was the wrong protocol. Centrelink had been relying on a separate protocol, originally written way back in 2004. Centrelink hadn’t updated the protocol before embarking on the dragnet in the second half of 2016; the most recent update is dated May 2017 and was not in place before April. The Department only published the 2004 and 2017 protocols on May 30 this year, in an unannounced update to the DHS website.


What of the Protocol itself? What does it tell us about what Centrelink is doing?

The answer, for those who know how to read such documents, is quite a lot. And what it reveals is a Department with fundamental problems in the way it designs and manages its data-driven compliance systems.


If you want a real insight into the design mentality of the robo-debt algorithm, the “risks” section of the Protocol is telling. This is meant to be the section where the program designers anticipate possible issues that might threaten the integrity of their roll out.

But what risks does the NDEIM protocol identify? Very few. The risks section states blithely that:

DHS Customers will be provided with the opportunity to verify the accuracy of the information before any compliance action is taken.

Once again, the blanket assumption is simply that welfare recipients will be able to correct any issues.

There is no discussion in the risks section on risks to privacy.

As one data scientist that looked at the protocol told New Matilda, “this was, in reality, a massively risky project.”


Centrelink had changed the rules. But they continued to insist nothing had changed. The Department of Human Services only published the 2017 update to the data matching protocol on May 30. Centrelink’s updated Non-Employment Income Data Matching protocol is scheduled to go into operation only later this year.


There was a revealing moment on the last day of the Senate Inquiry hearings, when Labor Senator Murray Watt asked Secretary Kathryn Campbell about Centrelink’s error rates.

“So there are 10,000 – what do we call them?” Watt responded.

“Initial clarification letters,” Campbell responded, with a thin smile.

It was all very jolly and jovial, as the Brigadier-Secretary with a $731,000 salary hopscotched her way around the truth with a series of bureaucratic neologisms. Jovial, that is, unless you paused for a moment to recall that at least two people have committed suicide after receiving such “initial clarification letters”.


This exchange between Watt and Campbell is instructive.

WATT: I remember that earlier on in this inquiry we were told that, for those 20,000 letters a week that were being sent out, there was about a 20 per cent error rate. Is that figure still right?

CAMPBELL: We have never agreed that that is an error rate, and I think you and I have discussed this on many occasions. I think the Ombudsman too has discussed it. In fact, does someone have what the Ombudsman said about the error rate?

CHAIR (Rachel Siewert): Let’s not repeat what the Ombudsman said. We do know what the Ombudsman said.

WATT: So they are not debt notices; they are initial clarification letters. If they are not errors, what are they?

CAMPBELL: They are initial clarification letters where the recipient or former recipient has been able to provide clarification, which means that there is no need to continue with the process.

WATT: All right. When I say ‘error’, what do you say?

CAMPBELL: I say there has been an initial clarification letter, the recipient or the former recipient has been able to clarify the information, and there is no requirement for them to continue in the process.

WATT: Okay. Let’s call them ‘clarifications’, then. Is that an acceptable term? Do you want to take that on notice?

CAMPBELL: We will take that on notice.

It seems that Campbell does not acknowledge that Centrelink is making any errors at all (“we have never agreed that that is an error rate”). It was nimble footwork, and in the moment Campbell seemed assured and composed. But if you stop to think about what she admitted, the exchange with Watt was devastating.

As QUT academic Monique Mann told New Matilda, “there are always errors in data sets”.


Campbell and her senior deputy Malisa Golightly were admitting that 20 per cent of Centrelink’s “initial clarification letters” were for fictitious debts. Debts that didn’t exist. Mistakes caused by Centrelink’s faulty data match.

Just don’t call them ‘errors’. They were, after all, just ‘clarifications.’

Think about that figure of 20 per cent. 20 per cent is one fifth. One in five of Centrelink’s data matches are wrong.

In fact, the true error rate is almost certainly higher. Data obtained from the Department of Human Services by Labor Senator Katy Gallagher shows the error rate for the robo-debt letters issued from the Australian Capital Territory between July 2016 and January 2017 was more than a quarter. The figures showed that of 2,591 debt recovery cases Centrelink opened from July to January in the ACT, 473 people were found owing no payment, while debt amounts were reduced for a further 195 – an error rate of 26 per cent.


Late in the final hearing, Murray Watt asked Kathryn Campbell if she had met with the Australian Council of Social Service, or ACOSS, Australia’s peak welfare body. No, Campbell replied, she hadn’t. The reason she gave tells you all you need to know: because ACOSS had called on the Department to halt the robo-debt juggernaut.

I have not, to my knowledge, received a request other than the one in January when it was a very heated environment, I think it is fair to say. I gave evidence that the discussions were not going to be quite productive at that stage, because it was all about, ‘Stop, don’t do it,’ rather than, ‘How can we improve going forward?’

In other words, Campbell refused to meet with ACOSS because she didn’t like what they had to say. Even for a senior Canberra mandarin, the arrogance was breathtaking.

“I do not always have the time to meet with everyone on every measure,” she concluded.

Observers like Senator Skye Kakoschke-Moore were surprised, to say the least. Kakoschke-Moore, a senator from the Nick Xenophon Team, has proved one of the most diligent members of the Senate Inquiry. On every indication, the robo-debt program was in crisis. Meeting with key stakeholders is crisis management 101.


ACOSS chief executive Cassandra Goldie told New Matilda that “in some ways the most disturbing part of this is the behavior of the government, in light of the deep distress and the groundswell of terrible experiences from people affected”.

“This has had devastating impacts on individuals, and it has shocked me that senior members of government were not alarmed by this themselves, and did not immediately call in the representatives of people affected to get to the bottom of what was going on,” Goldie said. “Instead we’ve had senior ministers essentially saying that this system is working, and we’re just dealing with ‘teething problems’ or needing improvements.”

“The system that we’re talking about here – the social security system – is actually the lifeline for people when they are not able to look after themselves, whether it’s because they’re sick, or looking after young kids, or caring for someone else,” Goldie continued. “This is about will they have enough money to keep a roof over their head… that’s what at stake here, it should be at the heart of the government’s concern.”


When the Tasmanian Council of Social Services’ Kym Goodes got to speak to the Senate Inquiry, she didn’t mince words.

“What the Australian government has done to Tasmanians in the last six months is almost incomprehensible,” she told the Senators.

It has been an erosion of trust, an erosion of human rights, an erosion of dignity and an erosion of the principles of good government decision-making. Unconscionable conduct is generally understood to mean conduct which is so harsh that it goes beyond good conscience. It is conduct that is particularly harsh or oppressive, conduct that is even more than unfair, and it must be against conscience as judged against the norms of our society. It was always highly likely that this system would be particularly harsh in Tasmania.

Goodes made the crucial observation: many welfare recipients simply don’t have the intellectual and emotional resources to deal with the heavy hand of Centrelink’s bureaucracy.

The chair of the Senate Inquiry is Western Australian Greens Senator Rachel Siewert. She was scathing of the Department of Human Services and Minister Alan Tudge’s bungling.


“There is no doubt that there is a power imbalance between the Commonwealth agency, which is the Department of Human Services, and an individual citizen. Many of the people that I have spoken to were overwhelmed by the enormity of what they were being required to prove, and they were crushed by that power imbalance.”

Siewert told New Matilda that it took 41 pages of web page print outs to describe a single interaction with the Online Compliance tool on MyGov.

“What Centrelink gave us is basically a handout which has 41 screens on it, I’ve literally got it here in my hand,” she said. “So you could have to go through 41 screens [to provide Centrelink with the correct income information], and some people have low literacy and numeracy skills.”

Kokoschke-Moore agrees. Centrelink’s byzantine bureaucracy is so bad, she said, that welfare recipients asked to provide extra information often have to lodge a freedom of information request for their own documents.

“We’ve heard that when people FOI, they are confronted with hundreds of pages of tables and acronyms that make no sense whatsoever; we asked the Secretary to look at the documents handed to us by a lawyer, [the Department]said it wouldn’t be a good use of their time, so what hope does that leave for you and me and ordinary citizens to understand those documents?”

Siewert points to the reversed onus of proof applied to welfare recipients caught in the coils of the Centrelink robot.

“The government continues to say the onus of proof is on the person to prove that the data is correct,” she said, “but essentially the government system is flawed, because it misidentifies a debt because the employer has not sent the periods of employment.”

“[That is] no fault of the person getting the letter. Because the data is not lining up, they get this letter. The government can do all the semantics they like to say the onus is not on the person… it is.”

After listening to weeks of harrowing testimony, Siewert has found the Senate Inquiry a draining experience.

“You come out of those hearings and you feel really drained. The evidence we hear is very distressing – hearing of people’s experiences and feeling their sense of powerlessness and despair.”


And yet, despite everything we’ve learned about the robo-debt disaster, the government is pushing ahead. In fact, the automated data-matching is expanding. The data trawl will expand to disability and aged pension recipients next.

New Matilda has already heard about one report – so far unconfirmed – of a $90,000 debt being issued against a disability pensioner. In the new financial year, the government will begin drug testing an estimated 5,000 welfare recipients.

One of the most fascinating aspects of robo-debt is the way it appears to be metastasizing across the federal government, pulling in agencies that no-one would have expected to be involved in the control and supervision of citizens.

Take Data61, the data science arm of the CSIRO. Data61 presents itself as “Australia’s leading digital research powerhouse”. But now we know the agency is also working for the Department of Human Services to help fix the robo-debt mess. The CSIRO is also going to be working on the drug testing of welfare recipients, as Senate Estimates testimony last month revealed.

To say that CSIRO and Data61 scientists are unhappy about this is an understatement. New Matilda has spoken to sources inside the agency that are deeply troubled by the development.

Industry news site InnovationAus has also picked up the disquiet. Journalist Denham Sadler reported in June that “the proposed Data61 involvement in identifying welfare recipients for drug testing has caused some anxiety among staff, with some experiencing an ‘existential crisis’ over the issue.”

The role of the CSIRO has traditionally been public-interest scientific research. Many inside CSIRO and Data61 question whether drug testing welfare recipients and tweaking the robo-debt algorithm is truly in the public interest.


And still the robo-debt tentacles spread. At every stage of the slow moving robo-debt disaster, both the government and the senior bureaucrats at the Department have covered up and denied. Time and again, in the face of graphic evidence, they have insisted that nothing is wrong.

The debt letters weren’t really debt letters, they claimed – even when they were handed over to debt collectors. The online compliance system wasn’t really a big change, they argued – despite ramping up a system that issued tens of thousands of new notices every month and that was budgeted to raise hundreds of millions of dollars from welfare recipients. There weren’t actually any errors, they said – and then admitted that the error rate was actually 20 per cent. The Ombudsman cleared the Minister and the Department, Alan Tudge claimed – when a close reading of his report showed anything but. Tudge keeps claiming “the system works”, even as appeals to the Administrative Appeals Tribunal have risen by 50 per cent.

Underlying the bluff denials is a belief amongst top department officials and the Minister himself that the welfare recipient is always to blame (even if they’ve done nothing wrong). As we heard repeatedly in the Senate Inquiry, if the Department is merely asking for more information, then the bureaucracy can’t be at fault.

The delusion is palpable. Tudge, Campbell and the top brass are completely divorced from the reality on the ground (or on the phones). Wracked by industrial action, intentionally expanding a broken data match program, and unable to answer the phone, Centrelink is an agency in crisis. But Tudge and Campbell can’t, or won’t, see it.

There are many unanswered questions. Many of the questions put to the Department by the Senate Inquiry and taken on notice still haven’t been responded to.


As the Inquiry’s chair Rachel Siewert asked at one point in our interview: “the risk assessment stuff – why wouldn’t you do that?”

Labor’s spokeswoman for Human Services, Linda Burney, calls it simply “a failure of public policy,” arguing that there was “no proper planning, no risk assessment or no consideration of its impacts at all.” 

“What is very clear is that there is a cultural problem at Centrelink, one that starts with the Minister and the Executive staff,” Burney wrote in an email. “It is a culture which only blames clients when things go wrong. Most of the staff I speak with want to help, but they don’t feel that they are allowed to.”

“There is no question that robo-debt was a failure of public policy, and ultimately that failure lies at the feet of Alan Tudge.”

“Let’s be clear what the Department was doing,” Victorian Legal Aid’s Dan Nicholson’s told New Matilda at the end of our interview.

“They had a significant rate in increase in the debt, they removed human oversight and they were using data matching assumptions that were fundamentally flawed.”

“There are very real human consequences to a very large number of people, who are having debts alleged against them, which are probably wrong.”