New Zealand’s Accident Compensation Corporation (ACC), has emerged as a key player in the government’s push to integrate artificial intelligence (AI) across public services, particularly through its groundbreaking trial of Microsoft’s generative AI tool, Copilot. Recent reports and internal evaluations highlight the substantial benefits this initiative has brought to ACC’s operations, raising interest among other government agencies looking to enhance efficiency and service delivery with AI.
In the past year, ACC became the first government agency to experiment with Copilot, an AI-driven productivity tool by Microsoft, testing it with a cohort of 300 users. The results were overwhelmingly positive: four-fifths of the trial participants reported significant improvements in both the quality and speed of their work. More notably, the AI tool has provided notable benefits to employees with conditions such as dyslexia, ADHD, and hearing impairments.
AI Experiment: Pioneering Public Sector Innovation
“No other government agencies have trialed this tool ahead of ACC,” stated an internal report obtained via the Official Information Act (OIA). This first-of-its-kind trial was a bold move, positioning ACC as a leader in public sector AI integration.
The initiative began in 2023, when ACC aimed to determine if AI could boost the speed and efficiency of their work processes. The corporation sought to understand whether these improvements could be realized without compromising the intellectual property (IP) or privacy of ACC clients—a crucial consideration in the public sector, where privacy concerns are paramount.
To mitigate potential risks, the AI was operated on a fixed dataset, segregated from ACC’s main client record systems. This meant that privacy risks were minimal, as the tool did not gather or collect client information nor learn from the queries posed to it. Consequently, the system didn’t store staff inputs or use them to enhance its knowledge, as would typically be the case in AI-driven tools designed for continuous learning.
“This work is … of great interest across the other government departments,” noted the ACC ethics panel’s report. As one of the few government entities with the technological maturity and approval to test AI tools, ACC’s experience is anticipated to serve as a benchmark for other agencies exploring similar solutions.
Boosting Productivity and Wellbeing: A Dual Win
The ACC trial was conducted in two phases—first with a small group of 25 staff members and later expanded to 300 employees. The results were not only impressive in terms of productivity but also in employee wellbeing. For example, the AI-supported tool helped employees with ADHD or hearing impairments focus more effectively during meetings and assisted staff with dyslexia in revising content. The integration of Copilot into daily workflows provided an enhanced environment where employees could thrive, reducing the cognitive load of tasks that might otherwise be cumbersome.
“Surveys done … have indicated that it has been extremely beneficial,” the report stated. The AI tool quickly became a highly sought-after resource within the corporation. ACC soon began spreading word of its success across the public sector and technology forums, encouraging others to explore similar implementations.
In parallel, the New Zealand government has started promoting AI’s use in other public sectors, particularly for education and healthcare, as well as introducing a rudimentary chatbot to assist businesses in navigating administrative processes.
Privacy and Ethical Considerations: Addressing Concerns
The use of AI tools in public services inevitably raises privacy and ethical concerns, particularly in an organization like ACC, where sensitive client information is regularly handled. ACC clients have historically expressed concerns about how their data is used and what information is stored by the corporation. However, ACC’s privacy impact assessment of Copilot provides some reassurance.
The assessment highlighted that while privacy risks—such as breaches, inaccuracies, or misuse of personal information—are always a concern, Copilot was deliberately designed not to integrate with ACC’s client record systems. This design choice was essential in mitigating the risk of personal data being collected, used, or disclosed by the AI tool. “The risks of breaching client personal information are mostly alleviated because [the AI] does not integrate with ACC’s authoritative client record systems,” the privacy impact assessment confirmed.
Moreover, the AI tool left an auditable trail, making it easier for any potential issues to be traced. There were also discussions within ACC’s ethics panel about the possibility of watermarking AI-generated documents. However, this idea was shelved when it was noted that documents spellchecked by traditional methods were not typically watermarked either.
ACC took further steps to ensure ethical AI use by providing all employees with thorough training on how to ask useful questions and properly review any outputs generated by the tool. As the use of AI expands within the public sector, ACC recognizes that the potential risks will also increase. “The primary risk is potential inherent AI bias,” the ethics panel report noted. The privacy team recommended ongoing monitoring, such as spot-checking outputs to ensure no personal client information is inadvertently entered into the system.
AI Limitations and Language Barriers
While AI tools like Copilot have significantly enhanced efficiency, they are not without their limitations. One key shortfall noted during the trials was the system’s inability to support te reo Māori, New Zealand’s indigenous language. Attempts at translations were found to be inadequate, leading the privacy impact assessment to conclude that “transliteration is a closer description of what the tool attempts to do.”
As a result, ACC’s Māori advisers, who participated in the trials, emphasized the importance of using qualified translators for any official translations rather than relying on AI tools, which might produce inaccurate or culturally insensitive outputs. ACC’s approach to balancing AI efficiency with cultural competence underscores its commitment to responsible innovation.
Costs and Contracts: The Financial Impact of AI Adoption
The financial outlay for the AI trials has also come under scrutiny. Documents obtained via the OIA reveal that ACC spent $70,000 on external contractors to assist with the trials. However, the corporation has refused to disclose the full cost of its year-long license deal with Microsoft for 300 Copilot users, citing commercial sensitivities.
Microsoft, which has seen substantial growth in New Zealand, was a key partner in these trials. The company reported revenues of $1.2 billion in 2023, a sharp increase from previous years. Microsoft has positioned Copilot as an “everyday AI companion,” embedding it into many of its popular products.
For ACC, the future financial implications of AI adoption remain unclear. While the initial trial has demonstrated clear benefits, questions remain about the long-term costs of scaling up AI use across the corporation and, more broadly, throughout the public sector.
AI in Public Services
ACC’s successful trial of Microsoft’s Copilot has sparked widespread interest across the public sector. As the first government agency to integrate generative AI into its workflow, ACC has set a precedent for other departments considering similar initiatives. The trial’s results indicate that AI can significantly enhance both productivity and employee wellbeing, especially for those with accessibility needs.
However, ACC’s experience also underscores the importance of ethical considerations, particularly in ensuring that privacy is maintained, that AI outputs are unbiased, and that cultural competence is respected. The trial has provided valuable insights into how AI can be responsibly used in public services, offering a model for other agencies to follow.
As more public sector organizations explore the potential of AI, the lessons learned from ACC’s pioneering work with Copilot will be critical in shaping future policies and practices. For now, ACC’s experience demonstrates that with the right safeguards and ethical oversight, AI has the potential to transform public services, making them more efficient, inclusive, and responsive to the needs of all citizens.
As ACC continues to refine its AI capabilities, it will no doubt remain at the forefront of the government’s AI revolution—one that could redefine the way New Zealand’s public services operate in the years to come.
