By Ross Heron | CEO, Australian Payroll Association
Voice of Confidence | Episode 2 of 7 | Josh McKenzie, Chief Technology Officer, ELMO Software
Ross Heron sits down with Josh McKenzie on decorative versus foundational AI, trust, human oversight and what buyers should really be asking for.
You're three slides into a software demo. The rostering screen looks familiar, the payslip looks familiar and then it appears in the corner of the slide: "Powered by AI." Everyone nods. Nobody in the room is quite sure what it actually does.
Over the last two or three years we've seen a lot of what I'd call decorative AI in our market. It looks good in the tagline but changes very little once the pay run starts. When I travel around Australia and ask heads of payroll whether anything in HR and payroll technology genuinely excites them, the overwhelming answer is no.
In the 2026 APA Payroll Industry Report, just 12.9% of respondents said they currently use AI in the payroll function.
So I asked someone who builds this technology to be straight with us. Josh McKenzie is Chief Technology Officer at ELMO Software. After more than 20 years in fintech, he joined ELMO just under three years ago. My opening question had three parts, which he spotted immediately: "Triple barreled question to kick us off. Why not?"
"Frankly, I wasn't super impressed"
His reason for joining was refreshingly candid. In fintech, he had worked with a number of HCM and payroll providers.
"Frankly, I wasn't super impressed, as compared to fintech. So when this opportunity came up, I saw a real opportunity for disruption."
It's not often you hear a technology leader say that about his own industry. It's also not far from what payroll leaders tell me.
The one question to ask in every demo
I asked Josh how a buyer can tell decorative AI from something more substantial. His test is simple: "Is this product possible without AI? If the answer to that question is yes, you're probably in the decorative AI world."
Decorative AI makes an existing process a little smoother, such as form filling, a chatbot or content generation. Josh says it still has its place. Foundational AI is different. "We're talking about AI where the product fundamentally can't exist without AI," he said.
His example was salary benchmarking. Traditional benchmarking data is out of date the moment you receive it and someone still has to map it to your roles, often by deciding a role is "a level three, because that's roughly where their salary is right now". A tool that reads the actual job description alongside industry, size and location was, in Josh's words, "fundamentally not possible five years ago".
An agent that checks every line
Then Josh asked a rhetorical question every payroll professional should sit with for a moment: how confident are you that you're paying your staff 100% correctly?
His answer is agentic AI, which he defines as AI that owns an outcome. Picture a payroll compliance agent that knows the employee, their award, their classification and where the award requires it, the tasks they did on a specific shift. It then works through the pay codes line by line and points out anomalies to the payroll team.
"I'm not suggesting that this would necessarily do a payroll person's job. I'm saying that it would make it so much easier, because practically doing that in an organisation is very, very difficult."
That's AI producing a useful outcome for the payroll manager, not decoration.
Confidently incorrect
When we survey our members about AI, the barriers that keep coming up are security and trust. In the 2026 report, the most common concerns about implementing AI in payroll were a lack of understanding about AI technology (67.9%) and privacy concerns (56.3%).
Josh split this in two. Security is the easier part: make sure your data stays within the jurisdiction you operate in and your provider holds recognised certifications, such as ISO 27001. Quality is harder. "AI is non deterministic by nature," he said. It won't always give the same answer twice, so ELMO keeps a human in the loop on all of its AI products.
"We will utilise the hell out of AI to make people's jobs easier but we're not deferring accountability and responsibility to the AI."
That puts the onus on the human. As Josh put it, "AI can be really good at giving you a confidently incorrect answer, particularly if it doesn't have the right information to make a judgement."
A change program, not a tech rollout
About 18 months ago, ELMO started asking what it could do today that it couldn't do yesterday. The answer reshaped roles, responsibilities and its operating model.
Josh's key lesson is to "consider this more of a change management program than a technology rollout". The surprise came from his own engineers. "Engineers are usually the first people to want to push the boundaries and try new things with AI. This wasn't the case, because there was a lot of psychological uncertainty in that space." His advice: involve people, listen and understand their concerns.
The token bill nobody budgeted for
I'll make a confession here. At APA we've been surprised by the cost of giving everyone access to tools like Claude or Copilot without really understanding the true cost of adoption. His view: "If you were to throw AI at people and say, use it, it can be a costly exercise."
Most tools charge by tokens and "I can ask the same question once poorly and once well and it will have two different cost profiles." His fix is to train people to ask well and build systems "curated to be cost efficient", rather than handing out unfettered access.
Show me the value
I asked Josh for one prediction he'd be comfortable being held to in 12 months' time. He believes buyers "will cease to be impressed by AI shiny features" and move straight to the questions that matter: what value does this give my organisation and what is the return on that investment?
For payroll leaders who tell me nothing excites them, that's a welcome shift. It was great to hear a technology provider putting value to the organisation and the end user front of mind.
What this means for you
Payroll professionals: Take Josh's test into your next demo and ask whether the product could exist without AI. Whatever the answer, the output is still yours to check.
Bookkeepers and accountants: Treat AI output as a first draft, not a final answer. Confidently incorrect is still incorrect, so check it against the source before it reaches a client.
HR professionals: Run AI adoption as a change program. Bring people in early, listen to their concerns and expect uncertainty, even from your most tech-savvy teams.
Finance professionals: Ask for the return on investment, not the feature list. Budget for usage costs and favour curated tools over open access for everyone.
Payroll compliance officers: Ask vendors where your data is stored, which certifications they hold and where the human sits in the loop before any AI touches a pay run.
Watch the full conversation
This is only part of the story. In the full episode, Josh also talks about:
-
How his leadership style has moved from leading from the front to choosing the right style for the situation, from a crisis to consensus building.
-
How a simple rostering need, like having someone with a first aid certificate on every shift, becomes a strategic workforce capability question that reaches into recruitment, learning and performance.
-
Why ELMO rebuilt a reporting product that customers said you "basically needed to be a data analyst" to use.
Watch Episode 2 of Voice of Confidence with Josh McKenzie
This is the second of seven Voice of Confidence conversations with leaders shaping Australian payroll. Sign up to Payroll News at austpayroll.com.au/newsletter-sign-up so you don’t miss the next one.
Voice of Confidence is the Australian Payroll Association’s leadership series, featuring conversations with the people shaping the future of payroll in Australia.
