Australia's tax authority is giving administrative employees access to Microsoft's Copilot for routine tasks, aiming to develop workforce capabilities that can eventually improve its central operations. Mark Sawade, CIO of the Australian Taxation Office (ATO), described the strategy at Gartner's annual IT Symposium on Monday, saying the agency wants to "build the muscle" by putting AI in workers' hands. The approach focuses on developing what Sawade calls "return on employee" rather than immediate efficiency gains or traditional ROI.
The ATO already has a decade of experience with machine learning and advanced analytics for purposes like fraud detection, but Sawade isn't deploying generative or agentic AI near those core processes yet. Instead, Copilot is being used to accelerate everyday activities. A recent internal hackathon demonstrated the potential: a COBOL developer who had never written a line of C# or worked in the .Net framework won the competition using GitHub Copilot. Sawade noted the agency operates technology platforms ranging from mainframe to Java. Paper-based tax returns in the 1990s took six hours to complete, compared to the 12 minutes taxpayers now spend using the ATO's app and online service.
According to Sawade, the agency doesn't anticipate that allowing admin staff to work with Copilot will produce obvious efficiency improvements or return on investment. "You will develop the muscle you need to learn how to improve business processes," he said. The hackathon winner showed "the ability to transcend some of the detail, how someone who understands core concepts can use generative AI to work in a domain they have never looked at before is really powerful." The CIO hopes the initiative will boost AI literacy and potentially develop skills that lead to AI-powered enhancements of core processes.
Sawade's long-term ambitions for AI include delivering similar time-saving improvements for businesses with more complex tax affairs, and he floated the idea of personal agents that remind taxpayers when annual returns are due and interact with other agents to prepare them. Identity remains the main obstacle to that vision, and the CIO isn't rushing to implement advanced AI systems, noting that the agency can learn from problems like token costs that have affected other organizations "without being far behind." Sawade advised that any AI project should start with clear outcomes and work backwards to increase automation. By treating basic AI tools as a training ground rather than a productivity solution, the ATO is betting that workforce familiarity will prove more valuable than immediate automation returns. The strategy reflects a calculated patience in an environment where many organizations are racing to deploy AI without building the internal capabilities to sustain it.