The interview was conducted by Ula Rutkowska (Senior Researcher, Apolitical) and edited by Christina Obolenskaya (MSc in International History, LSE and Communications Intern, Apolitical). Chris Fechner was named on Apolitical’s Government AI 100 2025.

Apolitical interviewed Chris Fechner, the Chief Executive Officer of Australia’s Digital Transformation Agency (DTA), about how he is leading the country’s AI implementation in the public sector. Fechner’s work is centred on improving government services with AI, enhancing digital accessibility and building trust in public sector technology.

Under Fechner’s guidance, Australia recently implemented a trial using generative AI tools to understand the transformative potential of AI for government operations. The key takeaways? Microsoft Copilot saved staff an average of one hour per day, with 40% of that time reinvested into learning and teamwork. In this article, Chris shares key takeaways and discusses how AI will reshape the public sector.

Q: Australia recently trialled generative AI tools. What were the key outcomes of this trial, and how do you see generative AI shaping the future of government services?

The DTA-led trial of Microsoft Copilot involved over 7,600 staff across 60+ government agencies. Key outcomes included:

  • Improved efficiency and productivity: Staff saved an average of one hour per day on administrative tasks.
  • Enhanced skill-building: 40% of saved time was spent on learning and team collaboration.

From an information and technology perspective, the trial highlighted the need to manage where and how information is stored and accessed as well as how our technology environments (especially legacy technology) are managed so that the AI capabilities perform as intended.

The trial's evaluation also showed that participants felt comfortable exploring novel capabilities that Copilot could bring to their various job functions, including programming and scripting, generating images and entire presentation decks, and searching and summarising databases. These unexpected uses of Copilot demonstrated its potential to enhance various aspects of government work beyond routine tasks.

However, the trial also showed that the higher the expertise level required for a specific task, the less likely the GenAI output would be equal to or better than the human expert's work.

Our challenge in moving to this future is how safely, quickly, and affordably we can change our workforce, processes, systems, and governance to ensure that we realise the intended benefits of using AI while avoiding the downside of unintended outcomes, like bias and hallucinations, and the possible harms that result.

__Q: What skills do public servants need to adopt technologies like AI, and what strategies do you use to help them prepare? __

In my opinion, for governments to start using and successfully deploying AI, we need to dispel the myth that AI is some magical, all-seeing, all-knowing capability that can immediately and permanently solve all existing problems.

In busting this myth, we need to break it down into five steps:

  1. AI is not just one thing. Just like many people have an abstract, amorphous understanding of what "in the cloud" is, we have done a similar thing with AI, especially with the hype surrounding Gen AI. We need to have decision-makers, implementers, and consumers of the various types of AI who know what different types of AI they are using and how they can be confident that they are accurate and fit for purpose.
  2. AI won't be suited to all government tasks. We must understand the appropriateness and risks of using AI for particular tasks or domains. For instance, using Generative AI to increase or decrease the payments to senior citizens of a pension automatically would not be a suitable purpose without some extraordinary controls. Meanwhile, using the specialised knowledge and detailed data from the ATO to identify possible fraud cases in the tax system using a narrow predictive AI would be a use-case that the public would probably expect.
  3. AI isn't accountable or responsible for the outcomes it produces—people are. In preparation for AI use, we need to identify, train, and make public who has the decision and use rights of AI in agencies to ensure we have the public's trust and confidence in using AI.
  4. AI in all its forms is still a tool, and data is its fuel. Government has long been a data collector, and AI is opening up multitudes of new opportunities and risks for using this data. To prepare governments for AI use, we need to focus on our data, whether that data be for training AI, grounding AI inference, or new data formed from using AI.
  5. Effective use of our AI assurance Framework — not just in development but throughout the AI lifecycle. By requiring all government agencies to follow the standard of the Transparency statements, we encourage trust. Through the Accountable Official standard, we demonstrate that we are proactively managing risk and collaborating across government to safely adopt AI for public good. Our policy also emphasises the need to build up our skills and understanding through education and training.

__Q: What can organisational and team leaders do to create a culture that’s confident and ready to adopt AI responsibly? __

An experimentation and growth mindset are important capabilities for government leaders. AI is not just a digital automation tool like the adoption of computers and systems from the 1980s. AI represents an opening of completely new possibilities and new ways of delivering government for the benefit of all people.

This is why, in 2019, the Australian Government introduced the eight AI Ethics principles — human, societal, and environmental wellbeing; human-centred values; fairness, privacy protection and security; reliability and safety; transparency and explainability; contestability; and accountability — to ensure that the AI solutions we use are safe, secure, and reliable and that the people who manage them are informed and accountable.

__Q: For those leading AI initiatives, what advice would you give on influencing leadership and ensuring AI adoption becomes a priority? __

Never in my long career in data and digital have I seen such a cacophony of fear, anticipation, and a good dose of FOMO by leaders in both Public and Private sectors. My advice is to take advantage of the interest by starting with an agreed approach to Governance (think Hippocratic Oath of "first do no harm.").

Next, establish a way to concentrate the wide-ranging interests of your different stakeholders so that the scarce AI skills can be applied in concert to experiment and test a few potential uses — something like a temporary task force.

Once you have established that AI has practical applications (not just hype) in your organisation, encourage your leaders to engage in a broad community to discover, validate, and prioritise the new opportunities that AI provides and to learn from the experiments and experiences of others.

Finally, this new technology world driven by AI will not be easy, and it won't be successful without missteps, so we need leaders who can accept some risk, try and fail, and keep coming back for more investments.

Q: What specific AI use cases do you see as the most promising for improving efficiency and effectiveness in government operations?

One significant opportunity lies in improving APS productivity through AI copilots. These tools augment the skills and knowledge of APS staff in various areas of work. For example, Microsoft's Copilot assists with tasks like drafting and revising documents, searching files for specific information, and organizing data. Our limited trial of these services showed an average time saving of about one hour per person per day, with further improvements expected through additional training.

GitHub Copilot, trained on billions of lines of code, provides real-time, context-aware coding suggestions. This means that less experienced APS developers can increase productivity in developing and testing software, while expert developers can have the AI copilot do the more routine work allowing them to focus on the most complex aspects of the software development task.

Specialist copilots tailored to government-specific domains, like social services or tax, are also on the horizon. Less experienced staff in APS knowledge worker areas will be able to ask and have their content questions answered in easy-to-understand language, allowing them to be more productive and lessening the amount of time that senior staff need to commit to knowledge transfer and training.

Another promising use case is the application of AI for predictive and analytical tasks, such as fraud detection. Governments handle vast volumes of payments, some involving sensitive and vulnerable cohorts, making the risk of fraudulent activity high. AI can sift through large datasets to identify anomalies or "signals in the noise," handing these cases over to APS for further investigation. This approach, leveraging government data and tailored AI models, has the potential to recover billions in lost funds.

__Q: Looking ahead, what is your vision for the role of digital transformation and AI in Australia’s government over the next five years? What’s next? __

Looking five years ahead in the field of AI is a very difficult task. With the explosive growth and advancement of capabilities in AI since the launch of ChatGPT, the boundaries of the possible are receding almost weekly.

My first part of the vision is that we will need to be very adept at dealing with a fast rate of change and this means balancing our AI investments with the likelihood that something will soon supersede them. This means ensuring a short time-to-value on these investments. It also means thinking much more about our technology and data architectures. Making sure that we do simple things like de-coupling our complex government technology solutions, that we utilise APIs to connect the various threads of our services, and that we carefully curate and manage our data.

We will also have AI agents in pretty much everything we do. Over time, these agents will take on more and more complex aspects of human support, and in some cases, this will move into fully autonomous AI.