Robodebt reparations: Govt initiates $721m refund after data-matching failures exposed

The Australian Government is set to issue $721 million in refunds to welfare recipients caught up in its abortive data-matching and automated debt recovery scheme, widely – and derisively – known as ‘Robodebt’.

Under the cloud of a class action lawsuit, the Federal Government has revealed it will initiate a nation-wide refund to more than 470,000 individuals caught up in the computer-generated debt recovery scheme. The reparative repayments, totalling $721 million, are expected to commence in July this year.

The algorithm-generated debt assessment program, launched by the then Department of Human Services (DHS) in 2016, appears to have miscalculated welfare recipients’ income – used as a basis for determining eligibility and payment amounts for certain government entitlements (for instance, the Age Pension and ‘Newstart’ allowance, now known as ‘Jobseeker’).

In making these automated assessments, discrepancies were erroneously detected between annualised income data sourced from the Australian Tax Office (ATO) and that of the income reported to Centrelink by welfare recipients.

Following public outcry, the robodebt scheme was put on ice in November last year. It was officially scrapped by the Government last week.

Services Australia, the successor to the DHS, has said it may refund all repayments made on debts raised wholly or partially from the ATO’s income data.

In defence of the scheme, Minister for Government Services Stuart Robert said the Government’s Income Compliance Program was designed to protect the welfare system and ensure the right payments were assessed and made.

He said Services Australia made $180 billion in payments a year, noting that, as a custodian of taxpayer funds, the Government needed to exercise diligence about monies paid.

Compliance & computer-generated assessments

The Government’s Income Compliance Program was supported by an Online Compliance Intervention (OCI) system launched four years ago.

The OCI’s algorithm – relying on inter-agency data matching – appears to have miscalculated the money “owed” to the Government in the 470,000 cases that involved Centrelink.

The algorithm used income averaging, based on ATO data, to calculate money paid or owed to Centrelink. It appears to have failed to account for a potential mismatch between fortnightly (and often variable) income reported by welfare recipients and the ATO’s annualised dataset.

Drawing on these computerised calculations, debt recovery letters were automatically drafted and mailed out to thousands of social services recipients.

The robodebt scheme has since sparked an ongoing Parliamentary inquiry, together with a class action lawsuit and a Federal Court ruling.

Earlier, Roger Clarke, Visiting Professor in Computer Science at the ANU, revealed that a lack of human oversight, a deliberate cost-saving element of the automated OCI system, was a critical flaw in the system’s design, leading to the mass-scale miscalculations.

While the OCI treated ATO income data as a total figure for the financial year, welfare benefits were based on income reported to Centrelink fortnightly – a figure that, Clarke noted, was not always consistent.

“Designing a new [OCI] system embodied an assumption that income reported to the tax agency was earned at a consistent rate over the relevant period of employment,” Clarke said.

Under the OCI scheme, where a material discrepancy was discovered between the ATO and Centrelink data, revealing a supposed “overpayment” by the welfare agency, a letter was automatically generated and sent to the welfare recipient.

The burden of proof ultimately fell on the recipient rather than the agency to demonstrate that an error was made.

“Recipients were required to demonstrate their innocence of the accusation, by providing evidence that showed that they had not been overpaid,” Clarke noted.

“The recipient was required to challenge the accusation, and to produce supporting documentation satisfactory to Centrelink.”

When the OCI system was implemented, he added, Centrelink stopped accepting copies of bank statements where payslips issued by the employer were unavailable.

“Only after sustained pressure did it accept that copies of bank statements could still be acceptable.”

Automated debt recovery

If robodebt recipient failed to challenge an accusation or provide evidence that Centrelink would accept, the DHS automatically declared that a debt existed and, in many cases, added a 10 per cent ‘recovery fee’ when instigating its debt collection measures.

“At no stage did Centrelink or its portfolio agency, [the then] Department of Human Services, discuss with ATO the nature of the ATO’s data or the design of the OCI system.

“Further, no consultations were undertaken with representatives of Centrelink’s operational staff who deal directly with benefit recipients, nor with advocacy organisations that represent recipients’ interests,” Clarke said.

A parliamentary inquiry into automated systems

The Government’s data-matching and automated notices continue to be investigated under a Parliamentary inquiry, with submissions closing in September.

The inquiry noted on its official website that data-matching and debt collection were not new for social services or tax entities.

The ATO, for example, has provided income information to the department for approximately 20 years, the inquiry committee noted.

The cross-agency data matching arrangements remained the same for decades. The difference lay in how the OCI’s algorithm used income information from the ATO to calculate the purported overpayments of income support.

Under its terms of reference, the Parliamentary inquiry said it would consider further evidence about the “legality of the online compliance program, as well as the impact and implementation of recent changes to the program, in a second phase of the inquiry.”

The inquiry will table an interim report in August 2020 addressing evidence received to date. It will also examine the ongoing impact of automated debt collection involving current and past income support recipients.

Data-matching techniques used by Centrelink – including limitations and uncertainties of data-matching techniques and error-handling processes – are being scrutinised.

The inquiry also assesses the handling of underpayment errors, including the number of payments identified and made through data-matching following an underpayment.

Use of debt collectors and the cost of the Government’s compliance program to date, including the projected and actual amount raised from the program, is also under investigation.