Queensland’s Office of State Revenue will apply a machine learning algorithm across all state tax lines from next month after a trial reduced land tax debts by five percent.
The agency is set to be the first public sector organisation worldwide to put the SAP Leonardo machine learning technology into production.
Though it started out as an internet of things (IoT) technology, Leonardo has now been expanded to cover essentially all emerging technologies that can plug into an SAP core, such as blockchain and artificial intelligence.
The Office of State Revenue (OSR) is currently undertaking a three year digital transformation program to make tax and revenue management more “customer-centric, digitally-enabled and data-driven.”
Within the transformation, OSR is pursuing a ‘tax insights’ project, initially as a proof-of-concept but now as a full production deployment.
Tax insights operates on a full SAP stack, though it is mostly powered by SAP customer retention (which includes Leonardo machine learning) and SAP’s HANA predictive analytics library.
The proof-of-concept was built and delivered in an eight-week sprint by OSR and an SAP innovation centre.
“The initial focus for the proof of concept was land tax debt, due to the high rates of payment default experienced for this tax line,” the office said.
“As part of the proof of concept, the application assessed more than 187 million data records of 97,000 taxpayers across seven years and was able to pinpoint and predict 71 percent of taxpayers that became debtors.”
Importantly, the algorithm brought in more land tax - an “estimated improvement of five percent in debt / revenue collection”, and that was enough to get the business case for full production over the line.
“The application built on SAP Leonardo machine learning will be progressively rolled out across OSR to all tax lines, commencing July 2018 with the migration of the proof of concept for land tax into production,” the office said.
OSR said that the trial was able to identify “high-risk events or influences” that could point to the “taxpayers most at risk of defaulting on their land tax payment.”
Importantly, the system could “distinguish between taxpayers who have the capacity to pay but choose not to and those taxpayers who are in genuine financial hardship and therefore require additional support.”
“Both types of taxpayers require different approaches and responses, and being able to know when to apply one response over the other is critical to a client-centric approach,” the office said.
“A more responsive and tailored approach to individual taxpayers will lead to more timely collection of revenue and reduction in debt across all tax lines, and additional revenue to fund essential services for Queenslanders.”
OSR said that the machine learning application had been designed to “propose the next best approach, payment suggestion, or debt management procedure to best support a taxpayer to meet their obligations.”
OSR said the tool could help staff to personalise payment plans for taxpayers who were most at risk of defaulting.
“We can also send them digital reminders that offer easy ways to pay and messages [to their smartphones] in advance to help them learn more,” the office said.
OSR said it was hopeful that a production use of the tool could ultimately “mean a reduction in the amount of money owed to Queensland.”
It presently oversees a $17 billion revenue base, which it has managed using an SAP tax and revenue management system since 2009.