Tristan Tan CA has expertise across multiple disciplines. He has worked in taxation for more than 10 years and is currently a commercial associate director with IFM Investors. He’s also studying his Juris Doctor (a graduate-level law degree that provide the necessary foundation for practising law in Australia) and is a chartered financial analyst (CFA) Level II candidate, studying the second level of the CFA program.

Born into a generation where tech is second nature, Tan has seamlessly integrated it into his professional practice. And he’s into AI. Really into it. Instead of listening to music while driving, he has wide-ranging, back-and-forth conversations with ChatGPT, expanding on his train of thought and following up interesting tangents.

Nurturing this interest, Tan contributed to the design of Empathetic AI’s tax copilot, Luna, built specifically for Australian professionals. Unlike generic tools, Luna is a domain-specific tool designed to help tax practitioners dramatically reduce research time by delivering accurate, legislation-backed answers to complex tax queries – complete with source references. Looking ahead, Tan sees Luna as a precursor to a new era of agentic AI – tools that don’t just respond to prompts but proactively assist professionals as autonomous, context-aware collaborators.

Given his background, Tan has a broad but deep understanding of how CAs can integrate AI into their practice, from completing basic tasks like organising emails and daily schedules, through to adopting AI as an integral part of their professional team.

Understanding what AI means for your practice

A big part of starting out with generative AI (genAI) is understanding how to get the most out of the software, says Tan. Techies call this prompt engineering but it could be better thought of as knowing how to ask the right questions in the right way to get the best response.

There’s no single correct way to learn how to ask the right questions, but as an experienced practitioner you need to leverage your deep professional expertise to frame precise, specific queries so what is being asked is airtight. As Tan says, while ChatGPT has a massive amount of knowledge, it lacks specific training on Australian tax legislation.

How does this work in practice?

“Someone who is inexperienced might just ask ChatGPT to tell them about a deduction, but the response that comes back would either be very general or not entirely reliable,” says Tan.

Using your professional expertise, the prompt could be framed along the lines of: ‘Tell me if section 8-1 will allow this deduction and whether there are any ATO taxation rulings on this matter. Focus on the application of the negative limbs.’ ChatGPT would then be able to provide a far more robust answer.

“So, my advice to CAs is to learn prompt engineering and use your experience to make the prompt as detailed as you can, so the AI can give you the best possible right answer,” says Tan.

He also emphasises that while genAI tools can be amazing helpers, the buck ultimately stops with the practitioner. AI can make mistakes or fabricate information – the tech industry calls them hallucinations – and practitioners need to use their professional knowledge to validate all responses by AI to ensure there are no errors. That’s also why it’s so critical for CAs to stay abreast of developments and changes in the tech.

“AI can provide a very compelling answer that, if you lacked experience, you might believe the answer simply because it’s so compelling,” says Tan.

Integrating AI into daily life

While ChatGPT is the best-known public genAI tool, many CAs will have access to other AI software such as Microsoft Copilot, which integrates OpenAI’s GPT technology within Microsoft 365 applications.

Tan’s employer uses Microsoft Copilot, which is embedded right across the Microsoft 365 software suite. This means it has access to everything from Outlook emails and PowerPoint through to files stored in SharePoint, Word and Excel. It’s a powerful combination, Tan says, because it lets CAs automate tasks across applications, deal with file management and generate new documents, spreadsheets and presentations, all using prompts.

“Start small, like getting Copilot to help you draft an email, and then move up the curve,” Tan suggests. “Being that person where you embrace it in baby steps will really build the passion.”

As with any habit, the more you use genAI, the easier it becomes – until, as Tan says, “you won’t know how you got by without it”.

Managing risk

Tan cautions that AI adoption within a firm will largely be governed by the organisation’s appetite for risk. This is because without specific configuration, public software like ChatGPT can store everything a user types into the prompt box and use it to further train the underlying foundation model. Consequently, this poses a risk that confidential information may enter the public domain.

For practitioners dealing with privileged information, this could be a real problem and one Tan views as a major shortcoming of incorporating public AI services into practice.

This is just one of the ethical challenges that AI can pose for finance professionals, so it’s helpful to have a thorough understanding of genAI’s strengths and weaknesses. (CA ANZ’s Ethical use of generative AI micro course delves into the details for Australian and New Zealand members.)

But these problems are solvable – at a cost. Public AI tools like ChatGPT can be configured not to retain chat history, although this isn’t the default setting, posing potential confidentiality risks. In contrast, proprietary solutions such as Microsoft Copilot include built-in security measures to mitigate these risks.

“Ultimately, AI will allow us CAs to deliver better services and be more strategic partners for our clients,” says Tan. “It will eliminate routine admin, allowing CAs to focus on higher level tasks needing human creativity and expertise.

“That’s my vision for the future.”