By Adrienne Silla | Head of Advisory, APA
AI is quickly becoming part of everyday payroll.
We are using it to draft emails, explain complex topics, summarise information, write procedures and increasingly, answer payroll questions.
And there is a lot to like about it. AI can save time. It can help us find information faster and can be available whenever we need it. But there is one question payroll professionals need to ask before accepting an AI-generated answer:
Where did that answer come from?
Because when you ask two different AI tools the same payroll question, you may not get the same answer. And in payroll, that difference matters.
We decided to ask two AI tools some real payroll questions. One was ChatGPT, a general-purpose AI tool. The other was Beryl, the Australian Payroll Association’s AI agent.
The results were interesting. Not because one tool is simply “good” and the other is “bad”. But because they demonstrate an important issue for payroll professionals: the information an AI tool has access to can influence the answer it gives you.
THE PAYROLL QUESTION
How do we calculate a redundancy for a previously full-time employee who is now part-time? For example, they served 8 years full-time and 9 months part-time.
CHATGPT SAYS:
The employee’s continuous service falls within the at least 8 years but less than 9 years band, which provides an entitlement of 16 weeks’ redundancy pay under the NES.
BERYL SAYS:
For an employee with 8 years and 9 months of continuous service, the minimum redundancy entitlement is 14 weeks.
SO, WHO’S RIGHT?
Beryl is correct on the NES redundancy scale.
The Fair Work Ombudsman currently states that an employee with at least 8 years but less than 9 years of continuous service is entitled to 14 weeks’ redundancy pay. The 16-week entitlement applies once the employee reaches at least 9 years but less than 10 years of service.
The interesting part of this example isn’t just that ChatGPT got the answer wrong. It is how confidently it got the answer wrong.
When I questioned where the information was sourced, ChatGPT responded with:
For someone who doesn’t have a strong understanding of Australian payroll, it would be very easy to accept that response and move on.
A payroll professional, however, should know enough to question it. And that’s an important lesson to keep in mind when using AI.
You still need payroll knowledge to know when an AI answer doesn’t look right.
THE PAYROLL QUESTION
When an employee moves from part-time to casual employment, what is the correct way to pay out and report annual leave? Do we need to terminate and rehire the employee to submit the payout as a termination, or would we cash out the annual leave with STP code C?
CHATGPT SAYS:
I would be cautious about simply using C for the part-time-to-casual change. The fact that money is being paid instead of leave being taken doesn’t automatically make it a “cash out in service.”
BERYL SAYS:
When transitioning an employee from part-time to casual, the employee’s accrued annual leave is paid out in full as a cash-out of leave in service.
SO, WHO’S RIGHT?
This example highlights a different issue.
ChatGPT hasn’t necessarily given a definitive wrong answer. Instead, it has been hesitant to provide a clear answer.
For a payroll professional trying to process a payroll, that can be just as frustrating.
Sometimes you don’t need a list of possibilities. You need to understand what the relevant Australian payroll guidance says and where you can verify it.
In this particular situation, APA has obtained written advice from the ATO supporting the treatment as a cash-out of leave in service.
The broader lesson is that annual leave cash-out rules need to be considered carefully against the employee’s award, enterprise agreement or other applicable employment conditions. Fair Work confirms that cashing out annual leave is subject to specific requirements, including the applicable industrial instrument and, generally, retaining at least four weeks of annual leave.
So again, the answer isn’t simply about which AI sounds more convincing.
It is about the information behind the answer and the knowledge of the person using it.
This is where Beryl is different from a general-purpose AI tool. Beryl has been developed specifically for Australian payroll professionals and is available 24/7.
Beryl’s responses are based on a defined set of payroll-focused sources:
• Australian Payroll Association Helpdesk tickets
• APA’s extensive knowledge base documentation
• Australian Taxation Office website
• Fair Work website
Beryl is continuously monitored by our SME's at APA. That deliberate focus is important. Australian payroll is a specialised environment. We are dealing with Australian legislation, Modern Awards, taxation, superannuation, leave, Fair Work requirements and specific payroll reporting obligations.
A general-purpose AI tool has access to a much broader body of information. That breadth can be useful for many things.
But when you’re asking a question about Australian payroll, more information isn’t necessarily better information.
You want relevant information. You want Australian information. And you want information that is appropriate to the payroll question you’re asking.
One of the reasons AI is so appealing is how easy it is to use. Ask a question and almost immediately, you receive a detailed answer. It can sound authoritative. It can use professional language. It can provide explanations.
But confidence isn’t the same thing as correctness. AI doesn’t necessarily know when the circumstances of your question are unusual. It may not know that the Award you’re dealing with has a particular provision. It may misunderstand the length of service. It may apply a rule from a different jurisdiction.
Or it may provide a general answer when your particular employee circumstances require a more detailed assessment.
And unless you understand payroll well enough to recognise those issues, you may never know.
That doesn't mean payroll professionals shouldn't use AI.
It means we need to use it intelligently.
AI can be an incredibly useful research and productivity tool, but the answer should be the starting point for professional judgement, not a replacement for it.
Beryl gives APA members a payroll-focused AI assistant available 24/7, helping payroll professionals find relevant information faster while still applying their own knowledge and judgement.
Because when it comes to payroll, the most convincing answer isn't necessarily the right one.