Nudge messaging using real-time analytics at tax time

Project summary

Everyone benefits when taxpayers get their tax returns right the first time they lodge them. This ensures the right amount of tax is collected so the government can deliver services for Australians. It saves the time and resources associated with post-lodgement reviews and audits for taxpayers and the Australian Taxation Office.

Tax gap estimates provide an understanding about whether the right amount of tax has been collected. The tax gap is the difference between what the ATO expects to collect and what would have been collected if every taxpayer was fully compliant with the law. Errors in work-related expenses are the largest contributor to the ‘individuals not in business income tax gap.’

Real-time nudging is one approach the ATO uses to help taxpayers fill out their tax returns correctly. It prompts taxpayers to check amounts in their tax returns and correct errors before they lodge.

Providing tailored assistance when needed makes the tax return process easier for taxpayers and contributes to a more efficient tax system. Supporting voluntary compliance through proactive and preventative activities is key to sustainably reducing the tax gap.

An individual completing their tax return online.
Image: Australian Taxation Office.

How AI contributes to the project

The ATO’s real-time nudging models run when a taxpayer ticks the declaration box to lodge their own tax return using myTax online. A message appears on the taxpayer’s screen if a claim is flagged by a model as abnormal, nudging them to review and self-adjust if appropriate.

The ATO uses predictive machine-learning models that compare what the taxpayer has claimed to other people who have similar attributes, such as income and occupation. Where amounts entered are significantly different from the predicted or data-matched amounts, the taxpayer receives a message prompting them to check the amount entered.

There are separate models for likelihood and consequence. These models are trained to detect patterns using tax return labels. They are retrained regularly to improve their accuracy.

An ATO team of data scientists undertakes AI modelling work, including the development and maintenance of the real-time analytics for nudge messaging models. Models use machine learning techniques of k-nearest neighbour analysis and neural networks.

Outcomes and next steps

Real-time analytics enabled over 712,000 real-time prompts to taxpayers to check amounts in their 2023ꟷ24 income tax returns for mistakes they may have made. This resulted in an estimated $92.6 million of revenue protected by averting incorrect returns.

The ATO used a randomised control trial to help evaluate the effectiveness of its real-time analytics nudge messaging. Results for the 2023-24 income year show that taxpayers who received a nudge message were 2.5 times more likely to review their return and 3 times more likely to adjust their data than those who did not receive a nudge message.

While real-time analytics nudge messaging has been successful, the ATO is cautious with its approach to expanding the program. The ATO recognises that adding more nudge messages may diminish their impact on taxpayer behaviour. It is therefore carefully considering how the program may be extended while striking the right balance.

The ATO continues to assess how advances in AI technologies and techniques can enable further improvements to the tax time experience and help support taxpayers to voluntarily comply with tax obligations.

Find out more

Australian Taxation Office (2025) How we use data and analytics, ATO website, accessed 25 July 2025.