Revenue NSW using AI to claw back overpaid and fraudulent COVID-19 business grants

Revenue NSW has assured the AI system it has implemented for compliance efforts is not anything like the federal government's bungled robo-debt system.
Written by Aimee Chanthadavong, Contributor

Revenue NSW has turned to using AI and data analytics to help it work through compliance efforts to assess whether people were overpaid or if there were fraud cases as a result of recent COVID-19 businesses grants that were provided by the state government.

"[We're] using a rule-based system where we can enhance irregular patterns in the data that we're seeing and quickly move them through a modelling process into investigation for fraud and compliance," Scott Johnston, Revenue NSW deputy secretary and State Revenue chief commissioner, told the audience at the 2021 digital.NSW event.

"A quarter of a million businesses received grants through this process. We can't look at all of them. There's extreme urgency to ensure that people got the money that they needed to but also to do it in the right way, and a lot of responsibility over the next six months and 12 or more will fall on us to effectively drive this this effort."

This will not be the first time Revenue NSW has turned to AI and data analytics.

The agency has been leveraging such technology since 2018. Initially, AI was introduced to help divert customers deemed as vulnerable away from "really strong enforcement action" due to owing debt to Revenue NSW and towards alternative options.

Through that function, Revenue NSW has diverted approximately 15,000 people a year from a pathway into something "more satisfactory and supportive", such as work development orders, community service, counselling, or rehabilitation, according to Johnston.

"Most of our customers, particularly in the fines and debt space, have a long history with Revenue NSW; they have significant debt that they owe and actually getting to a point where we can have a conversation and understand them, we can't do that for everyone," he said.

"So, if we can use our artificial intelligence from combining our data sets in different ways in predicting what vulnerability looks like, and through machine learning continuing to refine and improve that model, we only do better."

Johnston assured, however, the AI system it uses is not remotely close to the technology behind the bungled robo-debt system.

"Fundamentally, the way that we work is seeing that we have the ethical framework … and the values that we maintain is critical to us, that this augments our capability. This provides decision makers with more information about how they might follow official pathway within our legislation," he said.

"It doesn't put responsibility on a machine that we have no accountability for, and this is really critical to the way that we design and the step and the pace that we move forward."

"A really difficult part of our enforcement function that we have, and one that we have to consider very carefully, [is] how do we do it. Now, if just do it to any account where someone hasn't paid their debt, then we're going to cause significant stress in an area where there might have been a better way.


Editorial standards