Centrelink’s Data-Matching Fiasco, by Gerard McPhee

Policy background

Governments recognise that Australians in the main accept that there should be income-support payments to assist people in need at the time when they most need that support. There are various allowances historically known by names like sickness benefit, the dole, the invalid pension, the widow’s pension and so on. The newer names for these benefits are less descriptive and less helpful. There is also a consensus that these allowances should be neatly and closely targeted to those most in need. This is called ‘means testing’. Governments of both stripes are also sensitive to the prevalent social myth that there are many people who receive benefits who are not entitled to them because they are receiving other income. There is ongoing debate about the means to ensure that these people are detected and denied benefits they do not ‘deserve’ and made to repay any ‘overpayments’.

If one defines policy as a generalised statement of intent, then Australian governments have had a policy of stringent means testing to target resources to the most needy and to minimise expenditure. What could be wrong with the policy? It is long established, easily understood and broadly supported. How could such a sensible intent go wrong?

What is happening now?

In giving effect to this policy Centrelink has been issuing payments using an automated system to check reported income against Australian Tax Office (ATO) data. The rules for the two systems are very different. Centrelink is based on fortnightly reporting and the ATO makes tax calculations on an annual basis. Centrelink has been checking reported income against ATO records for years. There is a computer program that checks the data in the two systems and reports significant discrepancies. Forget the fancy language about ‘algorithms’; it is a matching process. Until recently, Centrelink was issuing approximately 20,000 notices a year for debts calculated under this system, with a manual review process to account for errors. Even so, at least 20 per cent of notices raised by this system were found to be incorrect in the sense that no debt existed. This is conceded by the government. Such an error rate in any other data-matching process would send a commercial enterprise out of business.

This system has changed. Now Centrelink has increased the rate at which it matches ATO records to Centrelink records to 20,000 a week, roughly a fifty-fold increase. This increase in volume automatically means that the number of errors will increase and that manual scrutiny will decrease. Fifty times more records checked means fifty times the chance of error. Fifty times more records checked requires fifty times the staff committed to the task or one-fiftieth the manual checking. Some civilians (IT speak for the non-technically literate) have thought that the wonders of computer science could help. Sorry, computers only do what they are instructed to do, which is quite different from what ministers and managers would like to have happen. As explained above, Centrelink is based on fortnightly reporting and the ATO makes tax calculations on an annual basis. Often the ATO has limited knowledge about income by the fortnight. Just look at your last group certificate.

Thus you can understand that, without a massive injection of resources and with gross ignorance about scalability (IT speak for more volume of calculations), there will be vast numbers of errors. In fact, fifty times more. These errors are going to directly affect the most vulnerable in our society because Centrelink means testing is designed to address exactly that group. They will have the greatest difficulty in responding to the error produced by the matching system. They cannot look to the Centrelink phone centre to help—it gives a busy signal for twenty-nine million calls a year. Seven million calls to Centrelink were abandoned last year. A fifty-fold increase in the error rate will only exacerbate this.

An example of how the error works is provided by Jason Heeris:

Let’s say you became unemployed on July 1st 2009. You claim a Centrelink benefit (e.g. Youth Allowance) for a few weeks, at $430 per fortnight… The rules for each fortnight are: if you earn $430 or below, you receive $430 from Centrelink. If you earn $430–$1173, your Centrelink payment is reduced by $0.50 for each $1 over $430. If you earn $1173 or more, you receive no Centrelink payment.

You eventually find a casual job, getting a couple of shifts some fortnights and none in other fortnights. In a good fortnight, you’ll earn $800. You report this income to Centrelink, and your payments are reduced to $245. In a bad fortnight, you’ll earn $0, so you receive your full payment of $430.

After a few months working at this job, you start to pick up many more shifts, more regularly. Now your income is $2000 per fortnight. You report this income, and since it’s over the threshold for Centrelink payments to cease, you receive no benefits any more.

When July 1st 2010 rolls around, you do your tax return. You’ve earnt $30,800 over the year (13 fortnights of $2000, 6 of $800), and you report it as well as your Centrelink payments of $4480. From July 1st 2010 you continue to earn above the threshold and claim no more payments from Centrelink. End of story, you’d think.


Six and a half years later, in January 2017, you receive a notice that you owe Centrelink $4480 plus interest.

The debt notice will be issued because Centrelink’s automated system will assume that you received your annual income in equal fortnightly payments. In the example above, those payments ($30,800 divided by twenty-six; $1185) would put you over the income threshold of $1170.

Having explained how the main error occurs, it is interesting to note other sources of error in the Centrelink process for matching. Centrelink does not match the employer name in the most precise and sensible way available to it. Once again, Jason Heeris:

To determine whether you received income from a source you didn’t report, Centrelink compare the employer you report to Centrelink against the employer you reported to the ATO. Centrelink do this comparison letter by letter, without accounting for variations in spelling, typos, punctuation, etc.

So if you worked for ‘So & So’s Icecream Parlour’, you might have written it exactly like that. But when they were entered into the ATO’s database they might have been entered as ‘SO AND SOS ICE CREAM PARLOUR’. Centrelink will regard this as two different employers, and assume you received income from a source you didn’t report. They will then recalculate the reduced benefits you should have been paid, and issue you with a debt notice for the difference.

Here’s a real example (thanks to @Info_Aus) of how a well-known company entered its own name into government tender systems:




PriceWaterhouse Coopers



PricewaterhouseCoopers – Australian Firm

PricewaterhouseCoopers ACT



Centrelink’s systems will regard all of these as completely different employers.

Note that all Australian businesses have an Australian business number—a unique numeric identifier—that both the ATO and Centrelink have on file. This would be a much more sensible way to match up businesses, but they don’t use it.

Other difficulties include notices being sent to your online MyGov account. This is suitable for many people, but recipients of Centrelink payments are exactly those with less education and access to technology and who are most likely to struggle with the digital world. Centrelink will also send notices by mail, which, if you have not changed your address since you last claimed, will work . But this could be six years ago. Have you moved in the last six years? Centrelink’s approach to debt recovery in general is a problem, including its outsourcing and the heavy-handed tactics used by the appointed collection agency. This matter is too big to be dealt with here, but major issues of social justice and due process are involved.

The fiasco is still in full flight as government ministers and their representatives defend the debt-recovery project against a storm of public criticism and face an inquiry by the Commonwealth Ombudsman. The outcome is still unknown, but it is already clear that the project is deeply flawed.

How did this happen?

The debacle was entirely predictable. I have already discussed how an increase in the volume of transactions will lead inevitably to increases in the gross error rate and demand on support services, such as the Centrelink call centre. This is simple arithmetic and would have been discussed formally, internally, but the minister must have chosen to proceed nonetheless. I know that the highly professional IT team at Centrelink would have briefed upwards about the differences between the two data sets and rules. They would also have explained the tight connection between complexity, resources, rate of change and outcome. Outcome is often manifest in the level of customer service. They would have explained that there are four elements in any system, such as the Centrelink system or the ATO system, or any other corporate information system. These elements of complexity, resources, rate of change and outcome are locked together, and a change in one will change the others.

The interaction between the four elements exhibits relationships such as the following.

  • A system with a high rate of change will require more resources than a stable system, for example, staff retraining or computer reprogramming.
  • A complex system will require more resources, for example, extended staff training or more programming, than a less complex system.
  • Reducing resources such as programming or staff will reduce outcome quality.
  • Changing a system without additional resources will reduce outcome quality.
  • Simplification can often reduce resource cost.
  • Stable systems reduce resources devoted to maintenance.

Centrelink’s systems are very complex and frequently changed. There is constant tuning of the system to manage new entitlements or restrictions. Any competent technical manager, including Centrelink’s, always knows this and would have tried to educate ministerial staff.

The central question is: why did the government choose to ignore the obvious and introduce the data-matching system anyway? There are at least two parts to the answer.

The first part is that ministers want to reduce Centrelink expenditure and have a deep-seated suspicion about a class of people ‘rorting the system’. The factual basis of that suspicion is not critical here. What is critical is that Cabinet has a predisposition to support any additional method to detect imagined ‘rorters’ or ‘undeserving’ poor people, and will do so at a policy level, not considering the consequences of added complexity or transactional volume to the systems it will call on.

The second part is that no one wants to tell the emperor about his new clothes or lack thereof. The non-technical managerial class in Canberra would have assured the minister that all would be well rather than accept any blame for floating the idea that a data-matching system could be used in this indiscriminate and unverified way. ‘It is literally blame aversion; it is not risk aversion’, according to Paul Shetler, who was handpicked by Malcolm Turnbull to head the government’s digital-transformation drive and was former digital-transformation office head. Shetler went on to say:

The justifications…are just another example of the culture of ‘good news’, reporting only good news up through the bureaucracy. That’s how it works in the bureaucracy. Bad news is not welcomed, and when bad news comes, they try to shift the blame.

This is not the first time that a major operation has failed. The recent failures of the tax system and the census system are illustrative. In each case, as Paul Shetler argues, the failures have not been ‘a crisis of IT’ but a ‘crisis of government’.

Some remedies

A new approach is needed for checking eligibility and means-test compliance. The world outside Canberra knows that simple rules are more easily followed. The bewildering complexity of the current categories and rules would promote poor compliance in the most honest and well intentioned of customers. This simplification drive should affect both Centrelink and the Australian Bureau of Statistics (ABS). For example, Centrelink staff should no longer retype information from one system into another. Centrelink is difficult and confusing for users to navigate online; it drives people to phone lines or forces them to visit a Centrelink office—both more expensive options.

Fixing the data in the Centrelink system by using Australian business numbers (ABNs) would enable Centrelink to check with the ATO without recourse to the mail-out madness currently under way. Asking the recipient to track his or her employer payments is ridiculously difficult in light of the data confusion in the ATO system, as illustrated by the PriceWaterhouseCoopers example. It would be enormously cheaper to automate the checking. The difficulties of tracking past employers (if still in business) is exacerbated by the rate of job change. ABS data show that, in the twelve months to February 2011, more than four million people changed their work status. While the average number of unemployed people in each month of 2011 was around 600,000, overall 1.7 million people looked for work at some time during the year. But, of these, fewer than 150,000 (8 per cent) spent the whole year looking for work.

Senior management in the public service must be educated about the fundamentals of IT governance. The ‘gang of four’ factors described above, not technical information, should form the basis of such education.

Clean, simple lines of accountability must be developed within governmental management. During the frequent government restructures IT systems often end up being passed from agency to agency. This multiplies complexity. Responsibility becomes diffuse and accountability is diluted with the move from department to department.

The most important principle of all has been succinctly put by Paul Shetler, as ‘Put users first, always’.

If we designed systems that were easy to understand and easy to use, the effect would be revolutionary. It is no use looking to the commercial IT industry for a user-focussed solution because the companies developing these programs (‘products’) are not user focussed in the least. They are customer focussed. The customer is the one who pays them. The user of the system is irrelevant so long as the customer is happy.

The work of the current IT industry produces systems with very high levels of complexity and low levels of usability because this is what the customers—that is, the senior bureaucrats and ministers—want. It is, in their minds, and consequently in their specifications, more important to search out the exceptions and the edge cases instead of doing what the business world has done for decades: understand that there will always be a level of component failure or bad debt or non-compliance and that the price of 100-per-cent compliance is ridiculous complexity, cost and human effort.

Canberra needs to start with the user experience, and as this fundamental premise drives up and back in the system, both dollar costs and outcomes will improve. This, however, is too simple for Canberra, and worse, entirely alien to the culture of insulated privilege in which governments operate.

About the author

Gerard Mcphee

No biography available.

More articles by Gerard Mcphee

Support Arena

Independent publications and critical thought are more important than ever. Arena has never relied on or received government funding. It has sustained its activities largely through the voluntary work and funding provided by editors and supporters. If Arena is to continue and to expand its readership, we need your support to do it.


Well done, Gerard. I did not fully understand the nature of the Centrelink debacle, until you explained it. I am also reminded of some of the software issued by the Victorian government, the most notorious being, of course, the education department’s “Ultranet”. It is not just Canberra that is having trouble coming to grips with the subtleties of policy and data management.

Comments closed