I was on a panel a few weeks ago hosted by The Access Group with leaders from ELMO and the Australian Payroll Association, talking about where AI is heading in HR and payroll. Somewhere in the middle of it I realised we had split into two conversations happening at once. One conversation was talking about capability. What AI can do now, what’s coming next, how fast it’s moving. The other was talking about control. Who’s actually using it, what data it’s touching, whether anyone would know if something went wrong.
It was great to have those two conversations taking place at once. In most organisations I work with, they’re not even in the same room.
That gap has a name now: shadow AI, but the name isn’t the point. The point is that employees are already using AI to get through their day, often without telling anyone, and most organisations have no real view of where or how. Not because people are hiding something. Because nobody built the guardrails before people started driving.
The answer isn’t to stop experimentation, and it isn’t to buy another platform and assume the vendor has solved the problem for you. It’s to build the organisational foundations that let AI be used safely, confidently and at scale. That’s the part nobody wants to slow down for, and it’s the part that determines whether any of the rest of it works.
AI Readiness is Not a Software Feature
One of the clients I’ve been working with recently introduced a new module from their HRIS vendor. It was a specific reporting module, allowing the client to create reports and analyse data from the platform using the AI tool, but nobody had defined how it should be used.
When people aren’t sure whether their HR or payroll platform even contains AI, it can look like a technology-awareness problem. Sometimes it is. More often it’s a sign of low adoption maturity. Organisations implement a platform without defining how its AI capabilities should be used, training people to use them, or setting standards for reviewing the outputs. The features are technically there. The organisation isn’t functionally ready to use them.
That matters because the value of AI doesn’t come from having access to it. It comes from knowing:
- Where AI can actually improve a task or a decision.
- What information the tool is allowed to touch.
- What a good output looks like.
- Who’s qualified to review it.
- When a human has to step in.
- Who’s accountable for the final result, regardless of how it was produced.
Buying an AI-enabled system doesn’t transfer accountability to the vendor. Once you implement the technology and act on its outputs, you own the decision, the process, and the risk. That’s not a technicality. It’s the whole argument this piece is making.
Six foundations for responsible AI adoption
You don’t need every open question about AI resolved before you move forward. You do need these foundations in place first.
1. Put a Qualified Human in the Loop
Human oversight is more than someone typing a prompt or switching an agent on. A person has to review the output, and that person needs enough subject-matter depth to actually know whether it’s right.
An IT professional can’t validate a complex payroll interpretation just because they understand the system it runs on. A junior team member shouldn’t be the only review point for a high-risk employee decision they don’t yet have the experience to assess.
A recent example I’ve encountered where the “human in the loop” existed on paper but the human wasn’t actually qualified to catch the error was with an award interpretation issue where the reviewer understood the software but not the SCHADS clause being applied. It meant the software was working as expected, but the output was incorrect.
The human in the loop has to understand the intended outcome, recognise when something’s wrong, and know when to escalate. If that person exists only to satisfy a governance checklist, you don’t have oversight.
2. Make Decisions Reconstructable
You should be able to explain how an AI-assisted output became a final decision. That means keeping enough of the source data, instructions, review steps and approvals to walk the process back afterwards.
If no one can explain how the answer was reached, the governance isn’t strong enough, no matter how sophisticated the technology behind it is. This is the part that gets skipped first under time pressure, and it’s the part you’ll wish you’d kept when something gets questioned six months later.
3. Establish Clean Data and a Clear Source of Truth
This is the foundation everyone wants to skip, because it’s not exciting and it’s not quick.
I worked with a community services provider who wanted the AI flagging roster patterns that might breach award conditions before they reached payroll. Whilst the instinct was right, unfortunately the rostering data wasn’t ready for it. It sat across multiple systems that didn’t agree with each other, and before the tool could produce anything trustworthy, we had to go back and do the unglamorous part first: cleaning the data, agreeing on a single source of truth, and reworking the tool itself before it could produce an output anyone could actually rely on.
The technology usually isn’t the blocker. The blocker is that the underlying data lives in three different systems that don’t agree with each other, and nobody can say with confidence which one is correct. AI doesn’t fix that. It processes it faster, and it will produce a confidently wrong answer at scale just as easily as a right one. In a SCHADS or aged care award environment, a confidently wrong answer about entitlements isn’t a minor inconvenience. It’s a compliance exposure, and it’s someone’s pay.
Before you move towards more autonomous or agentic AI, you need clean HR and payroll data, a defined source of truth, clear privacy settings, and a lawful basis for how information is collected and used. This isn’t housekeeping before the real work starts. It is the real work. Skip it, and everything you build on top of it inherits the same fragility.
4. Redesign Work, Not Just Individual Tasks
AI adoption can’t stay a collection of isolated efficiency experiments. You need to map work across roles and processes, then decide what should be automated, what should be AI-augmented, what stays human-led, and what needs to be protected by mandatory human judgment no matter how good the tool gets.
This is where HR should be leading, not catching up. Job architecture, role design and capability mapping are already part of our discipline. HR can help an organisation see where AI changes a task, where it changes the value of a role, and where it quietly creates a new decision point nobody’s assigned accountability for.
Without that mapping, you get pockets of automation with nobody tracking the effect on workload, hand-offs, skills or the actual experience of doing the job.
5. Address Psychosocial Risk and Build Trust
AI changes more than process. It changes how people understand their role, how secure they feel about their future, and how their contribution gets valued. I’ve sat in enough of these conversations to know that the technical rollout is rarely what people are actually anxious about.
Leaders need to explain what’s changing, involve employees in redesigning the work, and give people a credible path to build new capability. Roles should be adjusted deliberately, not quietly hollowed out by automation until someone’s left with an incoherent, intensified job and no one told them why.
Transparency matters here too. When AI influences an employee’s case, calculation or experience, they should know when and how it was used. That’s not an administrative footnote. It’s part of building the trust that makes the rest of this work.
6. Develop AI Judgement, Not Just Prompting Skills
AI capability gets reduced to “write a better prompt” more often than it should. That’s useful. It’s not sufficient, and it won’t stay useful for long either; prompt mechanics date quickly. What lasts is judgment: understanding what the system is doing, knowing whether an output is credible, knowing what to check, knowing when to intervene.
Leaders need new skills here too. Managing performance has traditionally meant looking at someone’s time, activity and output. Increasingly it also means assessing how effectively they use AI, how rigorously they verify what it produces, and whether they apply sound judgment to the result.
From Payroll Processing to Strategic Payroll Partnership
AI will change payroll work. That doesn’t make payroll expertise less important; it makes it more load-bearing.
As routine processing becomes more automated, payroll professionals get more room for planning, workforce cost forecasting, role budgeting and strategic advice. AI can help surface anomalies in rostering and time records before they reach payroll, checking patterns against workplace rules and prompting a qualified person to investigate the exceptions.
The shift is from processing every transaction to overseeing a more intelligent system, while keeping the expertise to challenge, interpret and verify what that system produces. The output of AI should always be treated as a draft until an accountable professional has reviewed it. Always.
Build the Foundations Before the Agents Arrive
Over the next two to three years, organisations will keep testing where AI can remove friction and give people time back. The bigger shift towards agentic AI, systems that complete multi-step work with real autonomy, will make today’s governance choices matter a lot more than they do right now.
Clean data, defined sources of truth, privacy controls, role clarity, review processes. These are what let organisations expand safely. Without them, more powerful AI just creates more powerful risk.
The goal was never to slow adoption down. It’s to make it intentional.
That’s the opportunity in front of HR and payroll leaders right now: map the work, protect human judgment, build capability, and make sure greater efficiency doesn’t come at the cost of accountability, trust, or the people doing the work. AI can do more of the work. The organisation still owns the outcome.
How Zest Can Help
Zest helps organisations prepare for AI-enabled work by bringing people, processes, technology and governance together. We can help you assess AI readiness, improve HR and payroll data foundations, redesign roles and workflows, select and implement fit-for-purpose systems, and build practical governance and capability frameworks.
Ready to move from disconnected AI experiments to a clear, responsible approach? Get in touch with Zest to start the conversation.