Deloitte's Dr Elea Wurth joins Future of Work: Australia to explain why Australian leaders must consider AI governance from the outset, not as an afterthought.
Bel Bennett
Partner Marketing Director
Workday APAC
Audio also available on Apple Podcasts and Spotify.
When AI first entered the business mainstream, much of the conversation was about possibility: faster decisions, productivity gains, new sources of value and new ways of working.
Now the conversation is getting practical. How do we scale AI with confidence? How do we determine our risk exposure? How do we manage token costs and measure ROI?
And perhaps most importantly, how do we make sure people trust and use our AI systems, right from the start?
That was the central theme in my conversation with Dr Elea Wurth, Deloitte's Lead Partner for Trustworthy AI in Asia Pacific, on the latest episode of Future of Work: Australia.
We started with a topic that hasn't always been in the spotlight: risk, governance and compliance. As Elea joked, "We're finally cool."
She has a point. AI has moved governance out of the back office and into the boardroom because it touches all areas of the organisation, impacting how leaders make decisions, deliver services and manage work.
That makes the design of AI programs critical. Governance must be built from the beginning, rather than treated as an afterthought.
Report
One of the strongest points Elea made was that many organisations are struggling to hold two ideas together at once.
On one hand, AI creates opportunity. It can help organisations move faster, improve productivity, lift customer and employee experiences, and unlock new sources of value. On the other, it introduces risk around transparency, bias, security, accountability, cost and trust.
Elea described this as the need to balance risk and reward. That balance is difficult when governance enters the conversation too late. She often sees leadership messages focused on innovation, speed, and scale, with “safely and responsibly” added afterwards.
In her words, that can become “a motherhood statement that is tacked on at the end”, rather than something embedded in “the culture and behaviours of an organisation.”
If responsible AI is only a phrase in a strategy deck, it will not guide decisions when teams are under pressure to move quickly. If governance is built after the technology is already in motion, it can feel like a blocker.
But when it's designed into the way an AI program is set up, it gives people clearer pathways to make decisions, manage risk and move into production.
That is the practical shift leaders need to make. AI governance needs to be present when value is defined, when use cases are prioritised, when data and systems are assessed, when success metrics are set, and when accountability is assigned.
Elea shared recent findings from Deloitte’s Banking on Trust research, which examined the state of AI governance in banking.
Elea’s message was clear: governance itself is not the problem. Poorly designed governance is.
The findings were telling because financial services is generally more mature than many sectors in risk management and regulatory discipline. Even so, Elea said some globally strategic important banks had “effectively no governance or very little ad hoc governance in place.”
For leaders outside banking, the point is not that financial services is the only sector facing this challenge. Quite the opposite.
If one of the more mature governance environments is still working through significant AI governance gaps, other industries should be careful not to assume that existing risk models are enough.
Every sector is facing its own version of this question. The specific risks may vary, but the need to design for trust, accountability and value is common.
Elea’s message was clear: governance itself is not the problem. Poorly designed governance is.
As she put it, “It’s not AI governance that gets in the way of innovation and speed. It is poorly designed or low maturity AI governance that gets in the way.”
That line stood out to me because it reframes the role of governance.
Good governance gives organisations the structure to move. It helps teams get from proof-of-concept into production. It gives leaders confidence that AI activity is aligned to strategy, operating within risk appetite and capable of delivering business value.
When I asked Elea how boards can become risk managers rather than risk avoiders, she described a “Goldilocks effect” in the boardroom.
At one end is FOMO: fear of missing out. These are the leaders who have read the headlines, heard the success stories and want to go all in on AI. The risk is that governance gets stripped back because it is seen as slowing progress.
That creates problems later. Elea explained that organisations can often demonstrate ROI in a proof-of-concept environment, but then struggle to move into production because “the frameworks and the structures are just not there.”
At the other end is FOJI: fear of jumping in. These leaders focus on the failures, incidents and risks. Their instinct is to wait until governance is “strict and strong and fortressed” before making meaningful moves.
That also carries a cost. As Elea said, “Every week or month that you’re not engaged in learning and investing and developing is a week or a month that you as an individual or your organisation is behind.”
Organisations need to be clear about what they are trying to achieve before they build, buy or scale AI.
The lesson for Australian leaders is that both extremes can create drag. Moving too fast without foundations makes it difficult to scale safely. Waiting for perfect certainty means the organisation falls behind in learning.
The stronger approach is to get the balance just right. Build governance into the program early, then keep refining it as capability, ambition and risk exposure change.
In the early days of generative AI, excitement often carried the business case. Elea has seen that change, particularly as Australian organisations begin looking for tangible returns.
Leaders are asking where the promised productivity gains, efficiency improvements and new revenue opportunities are showing up.
Our latest Workday research has some answers. While 61% of employees in ANZ say AI reduces the time tasks take, and 58% say AI has accelerated work in a productive way, when it comes to tasks that cross systems or departments, things fall apart.
In fact, more than 8 out of 10 employees report they spend significant time coordinating across teams, moving information between tools and reconciling conflicting data.
This hidden workload means the the ROI that leaders are looking for is difficult to find, likely because the organisation is not embedding AI into the trusted systems that already run their business.
Elea's advice was practical: organisations need to be clear about what they are trying to achieve before they build, buy or scale AI. That means defining the strategic objective, expected value, metrics that matter and reporting needed to track progress.
“You need to have designed that into your AI programs of work right from the outset,” she said. Elea also made the point that “the measure of trustworthy AI is tightly coupled with the success of your AI program.”
If employees and customers do not trust the system, they are less likely to use it. If governance is weak, the system may struggle to move efficiently from development into production.
It is fascinating to see how quickly the conversation about AI has move from worrying about low adoption, to worrying about the costs of runaway consumption.
Elea explained that generative AI has a natural brake on usage because it is driven by people. Someone has to prompt, read, think and respond. Agentic AI changes that.
AI agents can run autonomously, 24/7. This means token consumption can rise quickly, even without a clear view of whether that usage is creating value. As Elea warned, “Token consumption doesn’t necessarily mean success.”
Oversight is much easier to establish when accountability is considered at the start of the AI program.
This is another reason governance needs to be built in early. Leaders need visibility of where AI is being used, how much it costs, what value it creates and whether that use is aligned to strategy.
Workday's Flex Credits are a good example of getting full transparency into usage, to manage costs. Credits are consumed when agents or applicable platform capabilities are used in production, aligning costs to value.
Elea explained that the old 'human-in-the-loop' models do not operate at the speed and scale that agentic AI requires. But human accountability remains essential.
She was very clear: “AI can never be responsible or accountable for its actions. A human will always be accountable for an AI’s actions.”
People may not sit inside every individual decision, but they must oversee the system. In other words, they need to act as a 'human-on-the-loop'.
They need to understand how agents are operating, what guardrails are in place, what is being monitored and who is accountable when something goes wrong.
This oversight is much easier to establish when accountability is considered at the start of the AI program, rather than after systems are already operating at speed.
Towards the end of our conversation, I asked Elea what AI use in the boardroom might look like in five years. She said it all comes down to education.
"...only when company directors truly understand the technology that they're engaging with and their organisations are engaging with will they be able to truly uphold their responsibilities," she said.
The organisations in Australia that get this right will not be the ones that move fastest or wait the longest. They will be the ones that build governance into their AI programs from the outset, so that trust, accountability and value are part of the foundation, rather than an afterthought.
Report
A gap between effort and impact is the real ROI problem. Discover the solution.