evidoria

← Back to browse

Good practice

Robodebt — Australia's automated welfare debt system and Royal Commission

Australia · Canberra · See the Australia profile

Services Australia's 2016–2020 income-averaging algorithm wrongly raised AU$1.75 billion in welfare debts against ~470,000 recipients. A 2023 Royal Commission ruled the scheme unlawful, triggering a AU$1.2 billion settlement and 57 governance recommendations.

~470,000 notices
Automated debt notices issued (2016-2020)
1.75 AUD billion
Total debt raised (2016-2020)
1.2 AUD billion
Class-action settlement approved by Federal Court (November 2020)
1.872 AUD billion
Total settlement including legal costs (November 2020)
663 people
Welfare recipients who died shortly after receiving a debt notice (Royal Commission finding) (scheme period, per Royal Commission)
57 recommendations
Royal Commission recommendations issued (July 2023)
~7 %
Share of recipients for whom the income-averaging method was methodologically valid (methodology assessment)

Details

Maturity
Discontinued
Promoter
Services Australia (formerly Department of Human Services)
Period
2016–2020
Keywords
automated decision-making, welfare compliance, income averaging, data matching, Centrelink, cautionary case

Context

The Robodebt scheme (July 2016 – May 2020) was operated by Services Australia (then the Department of Human Services) to automatically detect alleged overpayments of social security benefits. It cross-referenced Centrelink beneficiary records with averaged annual income data from the Australian Taxation Office, dividing a recipient's annual declared income into fortnightly estimates to infer undeclared earnings — a methodology valid for only about 7% of welfare recipients but applied universally without human review.

Objectives

The scheme's stated purpose was to automatically detect and recover alleged welfare overpayments at scale, replacing case-by-case human review with an averaging algorithm applied across the entire caseload.

Activities

The algorithm issued approximately 470,000 automated debt notices totalling AU$1.75 billion. Recipients — many unemployed, disabled or experiencing financial hardship — bore the burden of proving the debts were wrong, often without access to the original income data used to calculate them.

Results

A Federal Court approved an AU$1.2 billion class-action settlement in November 2020 (AU$1.872 billion in total including legal costs); the government simultaneously wrote off AU$1.75 billion in debts. The Royal Commission into the Robodebt Scheme (August 2022 – July 2023) concluded the scheme was 'a costly failure of public administration in both human and economic terms', describing the automated mechanism as 'crude, cruel and neither fair nor legal'. Commissioners documented 663 welfare recipients who died in the period shortly after receiving automated debt notices, and made 57 recommendations on human oversight, automated decision-making law, inter-agency data governance and accountability.

Conclusions

The scheme was discontinued in May 2020 after the government conceded in court it was unlawful. Amnesty International cited Robodebt in a 2023 report on automated risk-profiling systems as a case study of harm at scale caused by removing human oversight, and the case is now widely used in AI ethics and public administration curricula as a cautionary lesson on opaque, human-free algorithmic enforcement.

Implementation
Indicative cost
Very high (> €5M) — AU$1.75 billion in debts raised and later written off; AU$1.2 billion (AU$1.872 billion including legal costs) class-action settlement approved in November 2020.
Time to results
Medium (1–3 years) — Operated July 2016-May 2020 (~4 years); the Royal Commission of inquiry ran August 2022-July 2023.
Staffing & skills
Services Australia (then the Department of Human Services) — operated the scheme, Royal Commission into the Robodebt Scheme — independent inquiry, August 2022-July 2023

Common failure modes

  • Income-averaging methodology valid for only about 7% of recipients but applied universally without human review
  • Burden of proof placed on recipients, often without access to the original income data used to calculate debts
  • Fully automated, human-free enforcement at national scale
  • 663 welfare recipients documented to have died shortly after receiving debt notices, per the Royal Commission
  • Scheme ultimately ruled unlawful in court
  • Cited by Amnesty International (2023) as a global example of automated risk-profiling systems breaching human rights

Where it fits

Governance type
centralised federal welfare-compliance system, fully automated with no human review — the governance model itself is the documented failure
Scale
national
Income level
high-income (Australia)

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful