When a company decides who to hire, promote or make redundant, how much of that decision should be left to artificial intelligence?
It's a question becoming increasingly relevant as Australian businesses embrace automation and restructure their workforces.
This year, National Australia Bank recently announced plans to cut up to 2,500 jobs, or around six per cent of its workforce, as part of a major restructure designed to simplify operations and improve productivity. Elsewhere, Atlassian cut around 1,600 jobs globally (including almost 500 in Australia), Salesforce slashed its customer service workforce by almost half, and an Australian startup called Lorikeet developed AI agents that can handle complex customer service tasks.
While companies aren't declaring that AI is making redundancy decisions, these mass changes raise broader questions about the role technology should play in managing people.
In new research, Macquarie Business School academic Dr Miles Yang says while removing people from HR decisions may make organisations more efficient in the short term, it's dangerously inflexible when circumstances change.
"AI can make HR processes faster, cheaper and more consistent, but managing people isn't simply about following a set of rules," Dr Yang says.
"When decisions affect someone's career, livelihood or wellbeing, context and human judgment matter enormously. Algorithms can process data, but they cannot grasp the nuanced realities of human lives."
Could AI decide who gets fired?
Artificial intelligence is already used to support recruitment, workforce planning and employee management.
But a new generation of technology, known as agentic AI, goes a step further. Rather than simply providing information, these systems can independently carry out tasks and implement decisions with limited human supervision.
In an HR department, that could eventually mean an AI system identifying candidates for recruitment, allocating shifts, monitoring performance or triggering employment-related processes based on predefined criteria.
For businesses, the appeal is obvious. AI doesn't get tired, can process enormous amounts of information and can apply the same rules to thousands of employees.
But according to Dr Yang, consistency isn't always the same as fairness – or good management.
"An automated system might apply a policy perfectly, but that doesn't mean it has made the right decision for the individual or the organisation," he says.
"A human manager might recognise that an employee's performance has been affected by illness, family circumstances or an unexpected crisis. That context can be difficult to capture in an automated process, and treating people as data points inevitably erodes trust."
![]()
The trap of removing humans
The research identifies what Dr Yang calls the efficiency trap.
As organisations increasingly automate HR processes, they can become so focused on consistent execution that they lose the flexibility needed to respond to unusual situations.
When a business faces an economic shock, a workplace dispute or an unexpected crisis, rigid processes can produce decisions that no longer make sense and hurt their bottom line in the long run.
Imagine an AI system designed to identify underperforming employees.
It could analyse productivity data, attendance and performance indicators to flag staff who fall below certain thresholds.
But what if an employee has recently taken on caring responsibilities, helped colleagues through a difficult period or performed valuable work that isn't captured by the system?
A human manager may recognise these circumstances and decide that a different response is appropriate. An automated process may simply continue towards its predetermined outcome.
"The danger is that organisations become very good at following their own processes, even when those processes are no longer appropriate," Dr Yang says.
"Human managers provide the flexibility to question a decision, consider competing priorities and make exceptions when circumstances demand it. Without that human discretion, an organisation loses its ability to adapt when crises strike."
The paper argues this human involvement acts as a buffer, helping organisations absorb unexpected challenges rather than respond mechanically.
Should AI have the final say?
Dr Yang says the answer isn't to remove AI from HR departments.
Routine administrative tasks, straightforward scheduling changes and other predictable processes can be well suited to automation.
But decisions involving recruitment, dismissal, workplace disputes and employee wellbeing require greater care, particularly when the circumstances are complex or the consequences significant.
The researchers propose a governance framework that matches the level of human oversight to the risks and uncertainty involved.
Crucially, managers need both the authority and confidence to challenge or override an automated decision.
"AI should support better decisions about people, not remove the responsibility to make those decisions thoughtfully," Dr Yang says.
"The more significant the consequences for an employee, the more important it is that someone can consider the circumstances beyond what the system has been programmed to measure. At the end of the day, leaders must retain the authority and the moral accountability for the choices that shape people's futures."