The businesses getting the greatest returns from AI aren’t deploying the most tools or building the most agents. Instead, they’re following a disciplined three-step approach focused on data quality, process improvement and return on investment, according to CPA Australia.

New research from the accounting body found 88 percent of Australian and New Zealand organisations now use AI, but only 11 percent have deeply integrated it into their operations, highlighting a widening gap between adoption and business impact.

“They have a process of moving to AI rather than jumping to an AI solution.”

Gavan Ord, CPA Australia business and investment lead, told iStart the most striking result from the survey is how few organisations are extracting meaningful value from their AI.

“AI adoption is widespread, but the level of usage is shallow,” he says. Many organisations remain in experimentation mode, using AI for relatively minor tasks such as drafting emails rather than embedding it into workflows, decision-making and operational processes.

For many businesses, the stumbling block is working out how to apply it in their business.

CPA Australia’s annual small business survey shows small businesses often lack the time, skills or access to advice needed to realise the full potential of digital technologies.

“There are some deeper issues around the capacity and capability of small business owners in particular to take advantage of artificial intelligence and other digital technologies,” he notes.

The result is a widening gap between businesses experimenting with AI and those using it to drive measurable business outcomes.

The wider Business Technology Report, covering 1,679 accounting and finance professionals across Australia, New Zealand, Hong Kong, Malaysia, Chinese Mainland and Singapore, found a strong correlation between profitability and deeper AI integration.

Across all markets surveyed, 63 percent of businesses reporting increased profitability said AI was widely used across several business functions or deeply embedded throughout the organisation, compared with 50 percent of businesses whose profitability remained unchanged or declined.

“The real benefits come when AI becomes embedded in day-to-day operations, helping businesses improve productivity, efficiency and decision-making.”

Three steps and the case for less

So what can be learned from those companies that are reaping the full benefits of AI?

Ord says rather than rushing to deploy new tools and agents, successful organisations focus first on business processes, data quality and workflow design.

“They have a process of moving to AI rather than jumping to an AI solution.”

The first step is improving the data and data quality. Second is reviewing processes.

“Often they’ll hear from staff ‘oh, we need an AI solution for this problem’ but if they look at their process, sometimes they can improve the process without an AI solution,” he says, noting AI will not make a bad process better.

The third step is assessing whether existing software already contains AI capabilities that have not been activated or fully utilised. Only then do they evaluate whether a new AI solution is actually required.

The approach contrasts sharply with organisations caught up in the current AI gold rush.

Ord cited the example of one business owner who accumulated nine separate AI subscriptions before stepping back and reviewing what they actually delivered. After analysing usage and costs, the owner realised two subscriptions could achieve the same outcome, significantly reducing complexity and expense.

The same lesson applies to AI agents.

While many organisations experimenting with AI are building hundreds, or even thousands, of agents, some of the most effective organisations are doing the opposite.

Ord points to a large business-to-business payments technology company which runs its finance function using just four AI agents. Rather than pursuing scale for its own sake, the company invested heavily in a small number of carefully designed agents that deliver specific business outcomes.

The difference is strategic focus.

Show me the ROI

Businesses generating real value are not measuring success by how many agents they have deployed. They are measuring the return those agents produce.

That return on investment discipline appears largely absent from many AI conversations today, where discussions often revolve around experimentation, proofs of concept and productivity wins.

“Businesses that are successful at using AI are much more focused on the return on investment,” Ord says. “When they go through the process of choosing an AI tool, they have normal return on investment metrics in there.”

Leadership itself is also emerging as a strong success factor, Ord says.

The organisations making the most progress have moved beyond asking whether employees are using AI. Instead, leaders are asking why AI wasn’t used.

“The leader, the business owner, the management is really important to encouraging technology uptake and overcoming sometimes staff barriers to using technology.”

Ord says leaders need to actively use AI, demonstrating its benefits, sharing successes and failures openly and encouraging employees to look for new applications.

Governance as a competitive differentiator

The survey also highlights another challenge confronting A/NZ organisations: Governance.

A quarter of A/NZ employees use AI tools with little or no oversight or guidance.

Ord argues governance should not be viewed simply as compliance or risk management. Well-governed organisations tend to perform better during periods of disruption and change. In an AI-driven environment, governance increasingly becomes a competitive advantage because it enables organisations to scale AI safely while maintaining trust and accountability.

There is also a broader competitive issue emerging.

He says the conversations he has with organisations in China – where there has been a ‘very strong government push to use AI for a number of years – are fundamentally different from those occurring in Australia and New Zealand.

“When we speak to businesses in China, they speak in the past tense of how they’ve applied AI. When we speak to businesses in Australia and New Zealand, they’re speaking in the future tense of what they might do with AI,” he says.

“AI has the potential to make a meaningful contribution to the productivity challenge facing Australia and New Zealand, but only if businesses invest in the capability, governance and processes needed to use it effectively,” Ord says.

Three priorities

His advice for local businesses is straightforward: Deepen AI integration beyond individual productivity gains, establish clear governance with human oversight, and train both employees and managers to use AI effectively.

On the wider question of government support, Ord says there needs to be a substantial increase in incentives for businesses to use AI effectively, along with greater investment in AI literacy training.

He points to Singapore’s S$750 million investment over the past four years to encourage AI adoption among around 150,000 businesses as an example of the scale required.

By comparison, digital support programs in Australia and New Zealand target only a few thousand businesses.

“You’ve got one country that is investing heavily in supporting their businesses and investing in AI and another [New Zealand] that believes if it’s beneficial for business then there’s no role for government in helping,” he says, adding of Australia: “The level of ambition of Australian business policy support is significantly less than in Singapore and that’s one reason Australian small businesses are well behind Asia in terms of using technology, not just AI.”

Technology policies rarely decide New Zealand elections. Yet alongside the debates over tax, health and housing, AI, digital infrastructure and data governance are becoming increasingly important to the economy – and political parties are beginning to articulate their differing views about how technology should shape New Zealand’s future.

“Used well, AI has huge benefits… but we cannot leave it to big international tech companies to decide how it shapes our future,”

The debate extends far beyond the headline grabbing AI and data centres to encompass the broader environment in which organisations adopt technology, invest in new systems, manage data and navigate regulation. And while there is broad agreement – always something of a miracle in election politics – that technology will be critical to improving productivity and economic growth, there is less agreement on how government should encourage that transformation and whether active regulation is required.

At the centre of the divide is the role government should play in enabling technological change.

National and ACT generally favour a lighter-touch approach focused on encouraging adoption and investment. Labour and the Greens place greater emphasis on regulation, oversight and public accountability. NZ First’s digital policies are framed primarily around sovereignty, privacy and national control of strategic digital assets, while the Opportunity Party, currently the potential kingmaker, has put forward one of the most technology-focused economic platforms in the campaign so far.

R&D incentives and tech adoption

One of the clearest differences for businesses lies in investment policy.

National’s Investment Boost allows businesses to immediately deduct 20 percent of the cost of most new assets, including technology, with normal depreciation applying to the remainder. The party has committed to retaining the scheme. ACT supported the policy as part of the coalition and has not proposed any changes. Labour says it would repeal Investment Boost, using a claimed $7.7 billion saving from the program to fund its election promises. But it’s promising to raise the low-value asset write-off threshold from $1,000 to $10,000 allowing eligible small businesses to immediately deduct the full cost of new tools and equipment from their tax bill.

NZ First is proposing accelerated depreciation measures including a 100 percent depreciation rate for business equipment purchases worth up to $20,000.

The differences become even more apparent when looking at how parties would encourage organisations to adopt new technologies.

National supports the existing Research and Development Tax Incentive and has foreshadowed further announcements in the area, while NZ First wants expanded R&D tax credits and a $50m program to match private-sector investment in research, data analytics and advanced manufacturing. It has also floated a lowered – 20 percent – tax rate for SMEs with turnover under $30 million.

Labour’s focus, meanwhile, is less on tax incentives and more on direct investment through public funding mechanisms designed to support innovative firms and direct investment and support mechanisms including a proposed New Zealand Future Fund, an AI small-business fund, startup support programs and an expansion of the AI advisory service launched under the current government.

Labour says returns from the New Zealand Future Fund, inspired by sovereign wealth models like Singapore’s, will be both financial and social – stronger communities, lower costs, more resilient industries and opportunities that keep talent and ideas in New Zealand.’ “That might include backing community renewable energy or high-tech startups.”

The Opportunity Party has adopted the most aggressive tech-investment position. It’s proposing expanding the R&D tax incentive so businesses can claim up to 25 percent tax credit on investments in AI, digital platforms, automation, plant and autonomous equipment. It also wants higher government R&D spending, stronger support for startup incubators and accelerators, and a cross-party accord designed to provide long-term certainty around innovation policy.

The AI and data centre debates

AI has become one of the most visible tech issues of the campaign, although the parties aren’t divided over whether it should be used. Instead, the debate is around how it should be governed.

Both Labour and the Green Party are on the interventionist side, arguing that while AI presents significant opportunities, it also presents risks that can’t be left entirely to market forces or overseas tech companies.

Labour has unveiled a dedicated AI action plan saying it will take an ‘active role in shaping how AI develops in New Zealand, rather than leaving it to a handful of big technology companies to decide’. It has proposed an ‘Office of AI’ within the Department of the Prime Minister and Cabinet, promising it will be established from existing resources, rather than being a new layer of bureaucracy. It would be responsible for coordinating AI policy across government, developing ‘proportionate’ standards for responsible AI use, and developing a long-term plan for the New Zealand workforce.

“Used well, AI has huge benefits… but we cannot leave it to big international tech companies to decide how it shapes our future,” Labour leader Chris Hipkins says.

The party has also proposed an online safety regulator, a copyright framework to ensure creators are paid when AI systems use their work, and support for businesses and schools adopting AI and digital technologies. That support includes a $2 million a year AI small business fund, extension of the AI Advisory pilot implemented by the National led government this year to $3 million a year and an AI in Education team to support schools, teachers and students. The Founder and Startup Support Programme would also receive a $4 million a year increase, the party has promised.

The Greens have adopted an even stronger regulatory position. Their recently announced AI policy would prohibit fully automated government decisions that materially affect people’s lives, establish public registers of AI systems used by government agencies, allow citizens to challenge AI-assisted decisions and introduce transparency requirements for AI-generated content. The party has also placed significant emphasis on data sovereignty, privacy protections and limits on surveillance technologies.

“We will take a harm-based approach to digital technology, including social media, AI, and quantum computing, policy that upholds Te Tiriti o Waitangi and protects people’s privacy and online safety,” the party says.

Echoing moves seen in Australia, the party says it will ‘encourage digital systems to allow anonymous use’, remove trackers from publicly funded digital services and limit ID verification for digital services outside government without strong security protections for ID documents.

National, meanwhile has taken a more adoption-focused approach. The government released New Zealand’s AI Strategy and Responsible AI Guidance and has committed $70 million over seven years to an AI research platform. ACT has argued that New Zealand should be a regulatory ‘fast follower’ favouring international alignment over extensive AI-specific regulation.

On the data centre front, National supports continued data centre investment but, in a move similar to that seen in Australia, says it would require new large-scale facilities to bring their own firmed power, applying an ‘additionality test’ so that data centre load is always matched by new generation sufficient to cover it.

Labour also wants rules requiring large data centres to secure their own renewable energy and cover their own infrastructure costs, saying Kiwi households shouldn’t be subsidising the power bills of overseas tech companies.

The Greens have gone further, proposing a one-year moratorium on consenting new AI-focused data centres larger than 50MW while rules covering energy use, water consumption, emissions and community impacts are developed.

NZ First supports sovereign data centres, but says they should not displace domestic energy users or create adverse environmental outcomes.

Sovereignty, social media and digital advertising levies

Questions of digital sovereignty cut across several party platforms.

NZ First has adopted a strong data-sovereignty position, proposing that personal information held by public agencies be stored on servers located in New Zealand. It also supports sovereign data centres and wants a right to privacy added to the Bill of Rights Act. The Greens support Māori data sovereignty frameworks and stronger privacy enforcement, while Labour’s AI policy similarly emphasises New Zealand control over data and decision-making.

Online safety and social media regulation represent another area of clear disagreement.

National’s proposed Online Safety Bill would ban under-16s from major social media platforms and impose substantial penalties on companies that fail to implement age-verification measures. Labour supports the proposal and also wants deepfake regulations, restrictions on ‘nudify’ applications and an independent online safety regulator.

ACT, NZ First and the Greens oppose the social media ban, though for different reasons. ACT argues targeted interventions are more effective, NZ First says responsibility should primarily sit with parents and has called plans for a ban ‘a naive solution to a genuine issue’, saying there is no method that would fulfil that intent without banning BPNs or the use of digital ID to enforce the law. Meanwhile, the Greens favour a Digital Duty of Care regime, algorithmic transparency and stronger platform accountability.

The Greens have also been talking tough when it comes to tax avoidance and offshoring of profits by multinational tech giants ‘such as Meta, Amazon and Microsoft’ promising to crack down on those behaviours. It says it will ‘make big tech companies pay their fair share by enforcing the five percent withholding tax rate on the profits they send offshore’.

NZ First meanwhile is looking to scrap the accredited working visa model and replace it with a shortage-led Skilled Work Visa with entry tied to a named occupation, region and demonstrated need, rather than an employer obtaining accreditation.

Even media policy has developed a technology dimension.

Despite growing concern about the market power of global digital platforms, no major party is proposing a digital advertising levy to support journalism. National continues to weigh options around the Fair Digital News Bargaining Bill, Labour has proposed a levy on international streaming services to support local content production, and the Greens have focused on tax avoidance and market dominance among multinational technology companies.

The varied policies come as the Tech Users Association of New Zealand (Tuanz) and Tech New Zealand call for a bipartisan, long-term national technology strategy to prevent policies from shifting with every three-year election cycle. “We can’t afford to reset our national direction every three years while our global competitors move ahead,” says TUANZ chief executive Craig Young. “What’s urgently needed is certainty, coordination and commitment across political lines to deliver a joined-up national strategy.”

For businesses investing in AI, cloud, automation and digital, the election is unlikely to determine whether technology adoption continues. It will, however, influence how that adoption is encouraged, how it is regulated, who pays for the supporting infrastructure and who ultimately controls the data and digital assets that increasingly underpin New Zealand’s economy.

A customer-facing chatbot makes promises a business is ultimately forced to honour. A cyberattack destroys millions of euros’ of refrigerated stock. Sensitive discussions that once stayed inside a boardroom are automatically transcribed and stored on multiple machines increasing the potential for discoverable documents to be hacked. OpenAI agents ‘hack’ Australian sites.

None of those were part of anybody’s AI strategy. Yet all have become real-world examples of how AI can create consequences organisations never intended, from the Air Canada chatbot that invented a refund policy and left the airline liable for honouring it, to this week’s attack on the refrigeration systems of hundreds of French retailers, they demonstrate how increasingly sophisticated technology can amplify operational risk.

“You can outsource the model, but you can’t outsource the accountability.”

For governance specialist John Keogh, they highlight the growing challenge for organisations embracing AI: The biggest AI risks are often the ones nobody anticipated.

While much of the AI debate remains trapped between utopian promises and doomsday predictions, Keogh, partner at Sydney based advisory firm Centre for AI Governance, believes the reality is far less dramatic, and potentially far more important for business leaders.

“Some people think it’s the best thing that’s ever going to happen to us. Some people think all our jobs are going to disappear. Some people think the world’s going to end in three years,” he says.

“The answer is somewhere in between and balance, and balancing innovation with protection, is key.”

Accelerating risks, rather than creating new ones

One of Keogh’s central arguments is that AI should be viewed less as a standalone risk and more as an accelerant.

“It makes things happen a lot faster,” he says. And that acceleration applies to both opportunities and vulnerabilities.

The Australian Finance Industry Association has estimated the value created by AI and financial services by 2035 across areas including employee productivity, process automation, personalised marketing and customer experience to be around AU$60 billion. But, Keogh notes, those areas also represent some of the biggest areas of exposure if things go wrong.

The result is a balancing act many organisations are still struggling with.

Move too slowly and risk being overtaken by competitors. Move too quickly and risk creating problems that are difficult, expensive or impossible to unwind.

“It’s about balancing innovation and protection,” Keogh says. “If companies hunker down and don’t do stuff, they’ll get left behind.

What worries Keogh most are not necessarily the risks companies already know about, but the secondary effects they haven’t yet considered.

One example is the growing use of AI meeting assistants, recording tools and automated summaries.

Organisations may see them as productivity enhancers, but they can also create entirely new information management challenges.

“If there are a number of people on a call, that means there’s a transcript on pretty much everyone’s machine,” he says. “If any of that gets hacked and something sensitive has been discussed, then it’s in the ether.”

Board meetings and steering committee discussions often contain highly confidential – and discoverable – material. In the rush to automate note-taking and record keeping, organisations may inadvertently be increasing the number of locations where sensitive information is stored.

He notes the example of an Air Canada chatbot which told a customer bereavement discounts were available. The man paid the full fare but when he tried to get the bereavement discount was told they didn’t exist. In the ensuing Civil Resolution Tribunal case, Air Canada argued that it could not be held responsible for its chatbot’s autonomous communications, claiming the bot was a ‘separate legal entity’ responsible for its own actions. The tribunal rejected that argument and Air Canada was found liable.

“A lot of the unintended consequences are if it’s not programmed properly and the right checks and monitoring is not in place, then things can be promised that shouldn’t be. So you’re really exposing the organisation to things happening that you hadn’t accounted for,” he says.

The challenge with AI stuff is that if you make a mistake in the normal course of events in financial services – a change of policy or change a setting or credit scoring or something like that – it takes a while to push through the system. What AI does is it accelerates everything. So if there’s something wrong, it happens at scale. And a lot of that stuff’s difficult to undo once it happens.”It’s not just bespoke solutions serving up challenges either. Keogh says with vendors deploying AI within their tools organisations need to really understand their supplier base – and the potential exposure it brings.

“Often a very small supplier by spend can have a very significant impact on your organisation because of access to data and so on.”

Obligations apply

Despite the hype around AI, Keogh believes many organisations remain stuck in what McKinsey has described as ‘prototype purgatory’ – running pilots and experiments without moving significantly into production.

Others are waiting for regulators to provide clearer guidance.

But regulators are increasingly making it clear that existing obligations still apply, regardless of whether AI is involved.

That means organisations cannot assume responsibility shifts to technology providers.

Keogh points to the well-publicised Air Canada chatbot case as evidence that organisations cannot assume accountability rests with the technology itself. After the chatbot provided a customer with inaccurate information, the airline ultimately remained responsible for the outcome. The incident has become one of the most cited examples of what can happen when organisations assume AI-generated responses are someone else’s problem.

“The regulators and the AI vendors have made it clear that the responsibility for AI sits with the companies who deploy it,” he says.

Or, as Keogh puts it: “You can outsource the model, but you can’t outsource the accountability.”

Three steps

Rather than slowing AI adoption, Keogh says organisations should focus on governance, visibility and preparedness.

His first recommendation is understanding exactly where AI is being used across the organisation and which systems are touching customer data.

Second, executive teams need a shared understanding of both the opportunities and the risks.

Technology, security, legal, finance and operational leaders should be aligned rather than pursuing separate agendas, he says.

Third, organisations should ensure AI systems are properly tested before deployment and can be turned off quickly if necessary.

That capability becomes critical when AI systems behave unexpectedly. Keogh again references the Air Canada chatbot as an example of why organisations need effective monitoring and the ability to intervene quickly when things go wrong. “

You need to be able to make sure you can switch things off,” he says.

The ability to shut down a problematic system may prove just as important as the ability to launch one.

Underlying all of those recommendations is explainability.

If leaders cannot clearly explain how an AI-driven decision was reached, they may struggle to defend it to customers, regulators or shareholders when issues arise.

“If you can’t explain it at the kitchen table in simple English, then you won’t be able to defend it if something goes wrong,” Keogh says.

When energy giant Petronas was preparing for the AI era it deleted half its apps.

While much of the corporate world is busy bolting AI onto already cluttered technology estates, Petronas took a different approach, grabbing the pruning shares and decommissioning 50 percent of its apps. The discipline saw the Malaysian energy giant slash tech costs and embed a value framework designed to stop projects before they became expensive digital clutter. It also formed the beginning of a three stage journey to being digital.

“We poured a lot of money into IoT, analytics and stuff like that, but all it could do is just predict a machinery failure better.”

Shaharuddin Mustapha, Petronas Digital chief executive officer, admits the company is still only in the second ‘horizon’ of its plan: digitally enabled but still on the path to the ‘AI era’ and ‘being digital’.

Learning to say no

Speaking at Gartner’s IT Symposium on the Gold Coast recently, Mustafa compared the job of a modern digital leader to a game of Minesweeper – but Minesweeper where someone else is moving the mines, changing the grid and rewriting the rules while you’re asleep.

For Petronas, the challenge is bigger than technology.

The company operates in an industry where energy security, geopolitics, regulation, resilience and sovereignty can affect decisions as much as software architecture. Technology leaders can no longer focus solely on technical risk, he says. They must be able to explain risk in business terms, understand the value being pursued and decide how much control they are prepared to relinquish to vendors, governments and partners.

The changing environment led Petronas to a simple conclusion: Not every technology request deserves a yes.

“One of the principles that we put in was that the IT or digital organisation needs to have the ability to say no,” Mustafa says.

That’s not always an easy sell.

Before taking over Petronas Digital two years ago, Mustafa worked in the business itself, pushing for more capability, more investment and faster delivery.

“I was one of those people,” he admits. “I have the budget, just go do it. I want the cheap, good, fast.”

Then he moved to the other side of the table.

Suddenly, the reality of finite resources became impossible to ignore.

“We can’t do everything,” he says. “We cannot fulfil every single order.

The company responded by introducing an enterprise value framework to assess proposed initiatives. The key question became straightforward: what value will this create?

If the answer wasn’t compelling enough, the project didn’t proceed.

“We say no when the initiative creates no value,” Mustafa says.

Technology teams saying “no” was initially a shock to parts of the organisation, but Petronas sees that discipline as essential preparation for the next phase of transformation.

“The ability to say no is also a means for us to clean up our ecosystem,” he said.

That clean-up was the decommissioning of half of its applications because they were no longer needed. The company also undertook an extensive ‘cost compression’ program that reduced tech spend by around 20 percent, not through a major transformation program but by eliminating unnecessary services and negotiating more fit-for-purpose arrangements.

“We just took stuff out that we didn’t need,” he says.

The struggle for enterprise-wide value

Mustafa says the company is now in its digitally enabled stage. “A lot of platformisation happened here with the ERP, procurement systems and all that.”

Mustafa says Petronas invested heavily in cloud, cybersecurity, IoT and analytics long before today’s AI boom. But he’s open: Some initiatives delivered strong outcomes. Others generated isolated benefits that never translated into enterprise-wide value.

Case in point: Predictive maintenance. While the technology improved the ability to predict machinery failures, scaling that capability into broader economic value proved more difficult.

“We poured a lot of money into IoT, analytics and stuff like that, but all it could do is just predict a machinery failure better.” It never got to the point of scaling it the economic PNL of an asset.

On the flip side, one success story was optimisation of Petronas’ end-to-end liquified natural gas value chain, which consistently delivered results because it improved multiple parts of the business simultaneously, from production through to shipping and customer delivery.

“We’ve had the data coming in from the upstream production, sequenced to the onshore LNG liquefaction plants and then sequenced to the shipping, the marketing, and creative people to the end customer.”

Other cases, such as the predictive maintenance, were very task and use case based and couldn’t scale much beyond the initial use case.

“There have been many cases of value, but we haven’t got one that really scales up and is really pervasive,” he says.

The company is now looking – cautiously – at the value agents and AI can provide. “

We want to solve things in terms of the value of autonomy as opposed to the value of automation,” Mustafa says.

From manpower to outcomes

The company has also restructured how it delivers technology projects.

Petronas moved away from buying project manpower and shifted to outcome-based delivery models, forcing teams to define expected business value, workflows, data requirements and outcomes before work begins. According to Mustafa, the approach delivered an immediate reduction of around 30 percent in development costs.

The philosophy reflects the same underlying principle that drove the application clean-up: Technology should be judged on value, not activity.

For Australian and New Zealand CIOs facing mounting pressure to demonstrate AI progress, that may be the most important lesson from Petronas’ experience. The company is not arguing against AI. In fact, it sees significant potential in the next wave of autonomous systems. But before organisations start building an AI future, Mustafa’s message is clear: figure out which parts of the old technology estate deserve to come with you.

Sometimes the smartest technology investment isn’t a new platform. It’s hitting delete.

AI agents have moved from pilots and proof-of-concept into the enterprise mainstream.

At least, that’s according to new research from SailPoint, which found 79 percent of organisations now run AI agents in production environments. The findings are drawn from a survey of 340 senior identity, IT, cybersecurity and risk leaders globally, and suggest autonomous software has become part of day to day business operations.

Security teams recognise the risk of over-permissioned agents, but legacy ticketing workflows prevent them from auditing and scoping non-human entitlements dynamically at scale.

But, despite their apparent ubiquity, only two percent of organisations have deployed purpose-built tools designed to govern and secure those digital workers who are making autonomous decisions, querying critical databases and executing transactions on behalf of the business.

Nam Lam, SailPoint ANZ group vice president, says Australian and New Zealand organisations are embracing AI at speed, but many are managing AI agents using controls originally designed for human users.

“Australian and New Zealand organisations are rapidly embracing AI to drive innovation and efficiency, which is a testament to the region’s agility,” Lam says.

“However, our latest report reveals that 79% of organisations globally are managing non-human identities including AI agents with legacy identity security controls. This isn’t just a technical gap, it’s a significant business risk that leaves the door wide open for breaches.”

The Horizons of Identity Security 2026-2027 report found AI agents now represent 22 percent of all non-human accounts, underlining how rapidly autonomous systems are becoming part of enterprise operations.

Yet most organisations are still relying on technology never designed for that purpose. SailPoint says 85 percent continue to use either legacy identity platforms or platform-native controls, such as on-premises directories or manually maintained roles, rather than systems developed specifically for cloud, machine and AI-agent identities.

Built for people, not AI agents

That gap becomes clearer when organisations compare how they manage people versus software.

For decades, identity-management systems were built around employees. Staff joined the organisation, changed roles and eventually left. Access rights could be reviewed and updated through established processes.

AI agents operate differently.

They can work continuously, access multiple systems simultaneously and take action without direct human intervention. As their numbers grow, organisations are discovering the controls built for people don’t always translate neatly to autonomous systems.

While 87 percent of respondents rated their human identity-management capabilities as capable or better, fewer than half, 43 percent, said they had mature processes for managing agentic access. More than half, 54 percent, remain at the lowest maturity level for AI-agent identity security.

The report suggests many organisations are still struggling with the basics.

Only 15 percent of respondents said they could provision non-human access in real time. Another 15 percent reported mature capabilities for detecting shadow AI deployments, while just 43 percent said they felt highly prepared to provide evidence during an AI-related audit inquiry.

Lam says those challenges are particularly relevant in Australia and New Zealand, where many organisations are still building foundational identity-management capabilities.

“In Asia Pacific, with 37% of organisations still at the earliest stage of identity maturity, the immediate priority for ANZ leaders must be to extend their identity security strategy to discover, govern, and protect every single identity; human and non-human, because when something goes wrong, the tough questions will come from the board and local regulators.”

Practical headaches

The survey found organisations are less concerned about AI hallucinations than more conventional governance and security issues.

Data leakage and exposure topped the list of concerns, cited by 35 percent of respondents. Excessive permissions and access-control issues followed at 30 percent, while 27 percent pointed to autonomous actions occurring without human oversight. Hallucinations were cited by 17 percent.

Those concerns are also creating practical headaches for security teams.

Excessive permissions and privilege creep emerged as the most common challenge in managing AI-agent identities, followed by visibility and discovery issues and the ability to monitor and audit agent activity.

“Security teams recognise the risk of over-permissioned agents, but legacy ticketing workflows prevent them from auditing and scoping non-human entitlements dynamically at scale.”

The report also highlights what it terms a ‘pronounced awareness gap’ and a significant disconnect between leadership awareness and operational execution. “Across multiple survey measures, broad sentiment questions yielded high optimism, while granular technical questions exposed critical gaps,” the report says.

“This is not simply a matter of leaders being overly optimistic; it reveals a deeper awareness gap. For many leaders, ‘identity security’ still primarily means securing the human workforce, and their confidence reflects years of investment in maturing those legacy processes. The data shows this confidence is misplaced in the agentic era, exposing a fundamental lack of awareness of what ‘good’ now looks like.

“It is a strategy-execution gap where the ambition to secure the autonomous enterprise has outpaced the operational ability, and even the vocabulary, to do so.”

“Anyone who is dependent on supply chains and the logistical risk [involved], needs to deploy agentic AI in order to reduce that risk. It’s as simple as that.”

It’s a bold statement, but according to Ilan Oshri, director of the Centre for Digital Enterprise at the University of Auckland Business School, local organisations are overlooking what could become agentic AI’s most valuable business application.

“Agentic AI in the supply chain should actually be positioned as a risk mitigation capability – that is where we’re going to see the biggest return.

While much of the discussion around agentic AI has focused on productivity gains, procurement automation and operational efficiency, Oshri argues businesses need to flip their thinking. For New Zealand and Australian organisations operating in an increasingly volatile global environment, the bigger opportunity may be using AI agents to continuously monitor, assess and respond to supply chain risk before disruptions reach the bottom line.

“Most companies think ‘let’s pilot agentic AI where we don’t have high risk. We’re saying you’ve got it wrong when it comes to procurement. You need to try it out,” he told iStart.

Oshri’s research into how agentic AI could transform procurement as part of global research for Boston Consulting Group, found organisations should start agentic AI deployments in areas with high transaction volumes and rich data environments, rather than low-risk experimental projects.

“Our findings are actually the opposite,” he says. “Where there are a high degree of transactions is where companies are actually supposed to start, simply because of the dependency on data in order to make agentic AI beneficial for the company.” That high data guides agentic AI to learn ‘really quickly’ reducing the level of errors significantly.

For local businesses however, Oshri believes the strongest use case is not procurement automation. Instead, he argues agentic AI should be viewed as a risk management capability.

“In the case of New Zealand, agentic AI in the supply chain should actually be positioned as a risk mitigation capability – that is where we’re going to see the biggest return, at least at this point in time.”

His argument reflects the realities of the smaller, geographically remote economies, heavily dependent on global trade, shipping routes and international suppliers.

From freight disruptions and geopolitical tensions to supplier issues and logistics bottlenecks, local organisations face risks that can emerge with little warning and quickly impact operations.

Using exporters as an example, he notes their need to get product to other countries. Freight costs are built into pricing, but what happens if emerging threats affect logistics, shipping costs or supply availability. Likewise, plenty of product is shipped to A/NZ to form part of products made here, and could benefit from similar agentic offerings.

“These kinds of agents basically do the scanning all the time and raise flags whenever they think a risk is going up.

Today, much of that monitoring is still performed manually or outsourced to specialist providers. Agentic AI could provide organisations with a constantly operating early-warning system capable of identifying risks – and taking action – before they become expensive disruptions.

“This is where possibly the highest benefit will be for companies.”

In the current environment, Oshri believes that capability is becoming increasingly important.

“Unfortunately, geopolitically, we are in the big surprises era at the moment,” he says.

While onboarding suppliers, managing contracts and automating routine procurement tasks all have value, he argues those applications are all ‘vanilla’ and relatively straightforward.

“You need to do that. But the big buck is the big surprises.”

Procurement should lead

The research also challenges traditional approaches to technology ownership. Historically, digital and IT teams have led major tech implementations across support functions. Agentic AI in procurement is different.

“We’re saying loud and clear that procurement needs to be the owner of the implementation of agentic AI,” Oshri says.

The reason, he argues, is simple: procurement teams understand the data they hold and the supplier relationships, business processes, contracts and operational risks in ways technology teams do not.

“Data scientists will make assumptions about the algorithm in terms of the ability to develop it and they will make assumptions about what data is needed in order to train it. That’s not the right approach to come up with an efficient AI-based solution.

“For this kind of technology, the approach should be completely different than what we have seen in ERP and what we have seen in RPA… It’s about do you understand what you have so the algorithm can actually behave according to the business process.

“It’s only procurement people who can answer that.”

An uncomfortable conversation about autonomy

Perhaps Oshri’s most controversial argument is that organisations also need to start thinking beyond human-supervised AI.

Current deployments typically are largely ‘human in command’, with some ‘human in the loop’. In either case people retain oversight of decisions and outcomes. Oshri believes many businesses are avoiding a more difficult discussion about autonomy.

“When we hear autonomous, I think many of us feel uncomfortable,” he says. “One reason for that is that we’re thinking if it is autonomous, that people just lost their jobs and it doesn’t sound right.”

He argues the transition towards greater autonomy is inevitable.

“As difficult as it is to think about, I think this is probably the future. We are moving towards that.”

Oshri says organisations should be honestly assessing which processes genuinely require human involvement and which could eventually be trusted to autonomous agents.

His view is that businesses often default to human-centred models because of concerns about the consequences, not necessarily because the technology is incapable.

“In many of the cases, we could have gone autonomous and we would have been actually providing outcomes that would not fall below the standards that we have built for human in the loop or human on the loop,” he says. “We are tilting towards the human-centred approach simply because we are struggling to cope with the consequences.”

It’s an argument likely to provoke debate among business leaders concerned about accountability, governance and workforce impacts. But Oshri believes the conversation is one organisations can no longer avoid.

For local businesses, the Oshri’s broader message is clear. The real value of agentic AI may not be found in making procurement processes a little faster or a little cheaper. Instead, it may lie in helping organisations anticipate disruption, manage uncertainty and build resilience in an era of increasingly unpredictable global supply chains.

And for companies exposed to those risks, Oshri’s advice is unequivocal: Deploy it.

The Australian Taxation Office is preparing for a world where individual’s personal agents will handle tax returns, talking to other agents via the government ecosystem – though CIO Mark Sawade admits there are still plenty of hurdles to overcome first, not least the issue of ‘Know your agent’.

Sawade says the ATO has spent decades improving how citizens interact with services through websites, apps and online portals, but AI may fundamentally change that model.

“If we’re not thinking about how we get there and the types of building blocks we’ll need to put in place, it’ll be an incredibly long time before we’ll be ready.”

Speaking at the recent Gartner IT Symposium, he pointed to the transformation of individual tax returns as an example. In the 1990s, taxpayers collected a TaxPack from the post office and typically spent seven to eight hours completing a return. Today, you can download the ATO app with MyID or access through MyGov and file a return for simple tax affairs in roughly 12 minutes.

The next leap, he suggests, may remove even that interaction.

“We’ve often thought about it through the lens of create an app or create a portal or, in the myGov sense, how do we bring together the set of federal government services that can be accessed together”

“[But] we will be in a world where we’ll be eventually talking to our personal agent saying, ‘it’s now mid-July, can you please just fill out my income tax return?’,” and the personal agent will talk to one or more other agents through the government ecosystem.

Sawade says the agency wants to reduce the cost and effort involved in business tax compliance, with a long-term vision of making tax obligations increasingly seamless and automated.

“We focused on the way the individual interacts with our government services to try and make that increasingly more seamless and that gets a whole range of better outcomes, both for individuals, taxpayers, and the economy broadly. Our focus now is how do we do that for business and how do we get that same type of step change so that the cost of business in terms of complying with tax obligations is equivalently reduced and tax just happens [as a] far more seamless and automated process.”

The future could see AI agents managing routine interactions, collecting information, preparing submissions and interacting with government systems without human involvement in every step.

It’s a similar story across at Services Australia, where deputy CEO and chief customer officer Jonathan Thorpe says agents could potentially do a ‘fairly good job’ of organising government information around specific individual’s circumstances.

“Certainly we’re seeing – and I think anyone would attest to this in most industries – that increasingly customers are experimenting with agents in their everyday life.

“Increasingly, we’re going to have to think about what that means in the context of government service delivery. I think we’ll start to see it – perhaps we already are – in the simple way people bring content together, organise government information around their personal circumstances, and potentially an agent could do a fairly good job of that.”

The challenge of know your agent

But both Thorpe and Sawade are clear that identity, accountability and trust are central problems still needing to be solved.

“Where it starts to run into a lot of trouble really quickly is proving who someone is, or know your agent,” Thorpe says. “At the moment we don’t have the plumbing necessary for agentic identity, particularly in the consumer and citizen space.

“I think we’ll solve it, but it’s not here yet,” he says, adding its something he thinks about ‘a lot’ in the context of MyGov.

Another ‘big challenge to solve’ is the issue of liability and responsibility. If an agent submits incorrect information or misunderstands instructions, clarity will be needed around who is accountable for the outcome.

“What happens if you set up an agent [but] you didn’t quite set the appropriate guardrails? What’s government’s responsibility in that interaction?”

Sawade too, is clear that we’re still a long way from agent to agent communications through government systems, pointing to identity as a key issue to solve.

“But if we’re not thinking about how we get there and the types of building blocks we’ll need to put in place, it’ll be an incredibly long time before we’ll be ready to have government services that are safely, securely and transparently interacted with in that way.”

Humans at the helm

Both emphasised the need to focus on outcomes, governance and accountability before introducing advanced automation.

At the ATO, that means balancing enthusiasm for AI against the reality that the organisation manages one of Australia’s largest repositories of personal and business financial information. Sawade says its approach is deliberately measured and designed around strong controls, governance and human accountability.

The ATO has ‘tweaked’ the commonly used concept of keeping a ‘human in the loop’ and instead talks about keeping a ‘human at the helm’ reinforcing the idea that a person must remain accountable for outcomes even as AI capabilities expand.

Services Australia is taking a similar position, with Thorpe saying he doesn’t see AI as a replacement for people: “I see AI as an opportunity to evolve that service and make it better, not replace it,” he says, noting agentic AI’s best role may be in assisting staff.

“I don’t think its in front of our staff, I think it’s actually behind it – how AI can support public servants to do a better job and better inform the community about what obligations, payments and services are available to them.”

That could take the form of an agent assist providing the right information as prompts to someone to connect them to the right service.

“Some of these things we’re looking at now as we transform our contact centre, but I’m not seeing it as something you’d have in front of a customer service representative, particularly in the types of things that we do.”

In a separate session, ATO assistant commissioner, strategy, design and architecture Jimmy Tzimopoulos noted that while agentic AI in particular can achieve a lot of user cases ‘just because you can, doesn’t mean you should’.

“When we really assess the different use cases, we might find that 90, 95 percent of them could be achieved through some form of re-engineering of process and then driving business automation, then maybe some machine learning and then a bit of agentic at the end or at a component of it. But there’s a tendency to say, I’ll just get it all done by agentic.

“And I think that’s the thing for us that we’re learning about.”

He noted the need to enable people to learn that while they can achieve things with agentic, you don’t necessarily have to look at agentic from start to finish. “Let’s look at how we can compartmentalise it to really deliver the right solution that considers the commercial impact as well.”

What happens when critical knowledge sits with one person in a business and that person leaves?

For The Warehouse Group a missing shoe buyer cost the company millions, with declining sales, weaker profitability, stock markdowns and lost market share as product assortments went largely unchanged for two years.

“Where you have the real risk in the organisation is your people and processes – and that’s where the unstructured data lives.”

The experience convinced group chief executive Mark Stirton that one of the biggest risks facing organisations isn’t a lack of data, but a lack of accessible knowledge, and it’s helped shape the Kiwi retailer’s thinking about data, knowledge and emerging AI technologies.

Speaking at the recent Snowflake World Tour Auckland, Stirton outlined how the company is exploring what he calls a ‘digital bench’ and an ‘agentic university’ as part of a broader effort to make organisational knowledge easier to access and less dependent on individual employees.

Stirton was blunt that the retail game is changing. “If you don’t think its changing, I think you might be under a rock. And when the game’s changing, you’ve got to decide what are the new rules? And what is the new playbook you need to run these rules against?”

He described what he sees as the next major shift in business. The first wave gave organisations access to manufacturing, infrastructure and eCommerce capabilities on demand. The next, he says is ‘pay-as-you-go intelligence’.

“When you think about tokens and all these models, you’re getting intelligence at scale, which has never happened in the history of man.”

The belief is influencing how the company is preparing for the future.

Stirton says he has challenged leaders across the business to rethink how their departments operate, saying ’30 percent of it must change’ as the organisation prepares for changing business and operating models.

The catalyst isn’t a new technology – not even AI. Instead it’s that recognition that critical expertise often sits inside people’s heads.

In the footwear example, the absence of a buyer did not simply create a vacancy, it triggered a chain reaction as assortments were not refreshed and decisions were delayed and effects flowed on to other areas of the team and business.

“It’s the contagion of small risks. They don’t look big in isolation but when you add them up they start to look like a really destructive force.”

Building a digital bench…

The challenge sits behind one of the more unusual concepts discussed during the presentation – that of a ‘digital bench’.

Using a rugby analogy, Grayston suggested many companies are trying to deliver the same outcomes despite effectively playing short-handed. Businesses expect results even when expertise is missing, teams are understaffed or key knowledge exists only with a handful of employees.

The digital bench is his attempt to address that problem.

Rather than relying on individual employees to hold critical information, Stirton wants organisational knowledge to become more accessible across the enterprise. The goal is not to simply store information, but to make expertise available when people need it.

That ambition is shaping the retailer’s technology foundations.

The Warehouse is working with Snowflake on semantic layers and vector-based capabilities, while also experimenting with Cortex to surface insights from existing dashboards and data assets. Stirton describes this work as the ‘unglamorous’ foundations required before organisations can take advantage of future capabilities.

He argues that structured enterprise data is only part of the picture.

While information stored in systems such as ERP platforms can be managed and analysed, many organisational risks sit elsewhere.

“But where you have the real risk in the organisation is your people and processes – and that’s where the unstructured data lives and where most of the gaps start to form when you lose a person.”

…And an agentic university

That thinking has given rise to an even bigger concept: what Stirton calls an ‘agentic university’.

Universities, he noted, create knowledge, preserve knowledge and transmit knowledge. In his view, businesses increasingly need to do the same thing.

He described a future where employees can work alongside knowledge layers and context layers that help connect information from across the organisation. Rather than expertise being concentrated within individuals, it would become easier to access and share.

“There would be abundance of knowledge and those knowledge gaps where there’s no training or there’s no parts where knowledge is transferred, that starts to be on tap for someone and then connected brains start to connect the whole organisation.

“So if I’m a buyer and I want to ask a question on my supply chain, suddenly you’ve actually got a knowledge layer – ‘your professors with your long-term knowledge’ and a context layer that now connects with each other.”

Stirton, who admits he’s ‘a bit of a dreamer and what I call a metaphor person’ referred to the concept as a corporate ‘second brain’. The idea is that organisational memory, historical decisions, operational context and expertise become easier to draw upon regardless of where they originate.

Stirton is not approaching the challenge as a company-wide technology rollout, saying he’s avoiding the ‘spray and pray’ approach that leads to burning tokens.

He says the retailer is focusing on innovators and early adopters – the 2.5 percent who ‘statistically will always move the organisation faster because they’re the ones who are braver, more courageous and curious’  arguing that change rarely happens evenly across an organisation. These groups, he believes, can help accelerate adoption and encourage new ways of working across the wider business.

“Adoption will always be uneven in technology,” he says.

“I think the people with the highest desire to have a growth mindset will start to use the technology and the knowledge layer and the context layer at much more rapid speed.”

The result is a view of AI that differs from many enterprise discussions focused on productivity gains or automation.

Instead, Grayston’s presentation centred on something more fundamental: knowledge.

Who owns it. Where it lives. What happens when it disappears.

The footwear example offered a costly reminder of the consequences when critical expertise resides with one person. The digital bench and agentic university are attempts to ensure that knowledge becomes an organisational asset rather than an individual one.

“3,064. That’s the number of Power BI reports we have in our environment Mitre 10.”

The comment from Nastassia Subritzky elicited plenty of laughs at Snowflake’s World Tour Auckland, but the Mitre 10 head of data and insights admits she was ‘pretty alarmed’ when she did the count during the hardware chain’s migration to Fabric.

“Our human users are also getting trained by our agents.”

“This reporting sprawl, plus the seemingly infinite number of spreadsheets makes it difficult because you’ve got all these different reports and spreadsheets and they all tend to have slightly different numbers, so you end up spending so much time trying to reconcile between all of them,” Subritzky says.

The problem arrived at the worst possible time, with the co-op already in the middle of a major ERP transformation, moving from individually customised store data warehouses to a centrally managed SAP platform and data warehouse.

“The move to a centrally managed data warehouse has naturally resulted in many cries for help with how to actually analyse data in this new SAP world,” she says.

Stores were struggling to work out where to go find the analysis they previously had available. The challenge was compounded by Mitre 10’s cooperative structure, where independently owned stores have their own approaches to performance and operations.

And then there was Excel.

“Turns out people get really attached to their Excel spreadsheets, especially during times of change,” she says.

Breaking free from report sprawl

Rather than create report number 3,065, the retailer decided to do something different.

The goal wasn’t to replace reporting, but to reduce dependence on it through AI agents built on curated datasets. Using Snowflake CoWork, the company created agents that draw from the same underlying data as Power BI, ensuring answers remain tied to a common source of truth.

The governance layer was critical.

Many organisations are discovering that generative AI amplifies existing data problems. Mitre 10 started with carefully controlled datasets, guardrails and verified queries to prevent users from generating yet another collection of disconnected answers. The retailer also maintained control over which datasets agents could access.

“With Snowflake Co-work, we have complete control over what datasets the agents have access to, and naturally to help with some reconciliation issues they use the same base tables as our Power BI workbooks.

“Everything naturally ties back together which has been a fantastic win for manual work reduction because I don’t have to send my engineers out to reconcile yet another Excel spreadsheet.”

Trent: Good tech skills, no business sense

The first AI agent, Trent, is the merchandising team’s ‘general assistant’. He was rolled out to the team as part of a two month deadline-driven migration from a legacy SQL reporting platform.

“This deadline doesn’t sound too bad until you consider how many workbooks someone can connect to a very old SQL reporting server over the course of a decade,” Subritzky quips. “We thought given they would have to get used to a new way of working anyway, why not just chuck them in the deep end and give them something really, really different to worry about?”

Trent helps with common tasks, including looking at supplier performance, undertaking basic rent reviews and looking at product performance by store.

But launching the agent was just the start.

“We got basic Trent up and running in absolutely no time at all, but he was very much like an intern. Really good technical skills, but absolutely no business knowledge whatsoever.”

What followed looks remarkably similar to onboarding a new analyst.

The team reviewed logs, refined prompts, built verified queries and taught the agent how the business actually worked. Instructions were added to standardise responses, eliminate exaggeration and ensure consistent analytical outputs.

Trent’s creator and Subritzky reviewed logs created from the merchandising team’s enthusiastic testing of Trent, turning successful requests into verified queries. There was some fun along the way too, with easter eggs added for testers.

The effort paid off.

“Trent had gone from being a really enthusiastic intern to a competent junior analyst who sometimes surprises us with some quite cool solutions.”

Today the agent is embedded in the merch teams daily workflow and has helped staff focus their analysis rather than build ever-larger spreadsheets.

Agents training humans

The store pilot provided more insights.

Initially, staff asked relatively simple questions. Over time, as confidence grew and the results were verified, the complexity of those questions increased. Store operators moved from asking basic sales questions to exploring inventory optimisation, promotional effectiveness and slow-moving stock analysis.

One example involved a store asking which products should be included in a winter clearance programme for slow and obsolete stock. Instead of simply producing raw figures, the agent asked clarifying questions, identified priority categories and recommended potential actions.

“What we were finding here was they could actually genuinely chat away to it because my team had done all the hard work up front to get all of those table joins going, all of the Mitre 10 context, all the odd little language that you end up having in business.”

Just as interesting was what happened to the users.

Agents were instructed to push back on vague or incomplete requests and ask for more information when necessary.

“In the interest of the cost, the compute time, and user experience, we added in explicit instructions about what sort of prompts the agent should go back to the user with if they didn’t get enough information.”

If users keep asking ambiguous questions, the agent keeps prompting them for more information.

Over time, users learned to write better questions.

“So our human users are also getting trained by our agents.”

That improvement in data literacy surfaced elsewhere as well. Because the platform could return the SQL underpinning its answers, analysts began developing a stronger understanding of how data tables connected and how metrics were calculated. The result was not simply more automation, but more capable users.

Meanwhile, executives discovered a different benefit.

With access to mobile capabilities, senior leaders began using the agents directly during meetings. Questions that previously required analyst support could be answered immediately. In stores, managers started interrogating data from the shop floor, investigating anomalies as they encountered them rather than waiting to return to a laptop.

Most importantly, the initiative appears to be delivering against its original objective. During the pilots, stores participating in the programme requested fewer new Power BI reports and modifications. Users were finding answers themselves rather than generating more reporting demand.

As Mitre 10 looks ahead, the challenge is ensuring history doesn’t repeat itself.

The retailer is already receiving requests for additional AI agents and is wary of replacing one form of sprawl with another. “We’re working out how we can make the existing analysts flexible, what sort of combinations we can use so that we can reuse a lot of what we’ve got and only do new agents when it really makes sense or is a really separate use case.”

After all, nobody wants to spend next year cleaning up 3,064 agents.

Australia has taken its strongest step yet towards taxing revenue generated by offshore software providers, with the Australian Taxation Office expanding the scope of royalty withholding tax rules covering software and intellectual property arrangements.

The move was quickly followed by New Zealand’s equivalent, Inland Revenue, releasing draft guidance on how software payments, cloud services and SaaS subscriptions should be treated for tax purposes.

“The ATO takes a broader interpretation of when a royalty may arise in cloud-computing context.”

But while both countries are trying to update rules written long before the cloud era, they appear to be taking different approaches.

Under the ATO’s new ruling, TR 2026/2, some software and SaaS payments may be classified as royalties rather than ordinary commercial transactions, exposing a wider range of cross-border software arrangements to royalty withholding tax of between five percent and 15 percent. It’s a ruling aimed primarily at multinational digital giants such as Amazon, Microsoft, Meta, Google and Netflix, and distributors who structure intellectual property and revenue flows across multiple jurisdictions, and will see income earned from cloud and streaming services become taxable in Australia for the first time beyond GST.

By contrast, the Inland Revenue draft guidance, issued earlier this month and updating a framework dating back to 2003, largely maintains the view that most straightforward cloud software subscriptions do not give rise to royalty payments because customers are buying access to software, not the right to exploit the underlying IP. Consultation is currently underway on the draft interpretation guideline PUB00266.

“The ATO takes a broader interpretation of when a royalty may arise in cloud-computing context,” Deloitte New Zealand noted in a recent tax alert.

In both countries, global tech giants generate substantial revenue – billions in Australia’s case particularly – from local customers, yet the profits ultimately taxed in the country have often been significantly more modest after accounting for distribution and service fees and other cross-border payments within their corporate structures.

Chasing profits, not GST

For years, governments have focused on collecting GST from foreign suppliers of digital services. Australia and New Zealand both require many offshore providers to collect GST on products ranging from streaming subscriptions and digital advertising to cloud services and software licences.

GST, however, was only ever part of the solution.

When businesses buy cloud hosting, software subscriptions or online advertising, the GST paid is generally reclaimed through the normal input tax process. Governments collect the tax, but much of it is ultimately returned. The larger question has been how to tax the profits generated from those transactions when the underlying intellectual property is owned offshore.

That’s a debate that has simmered for years as the global tech giants generated billions in revenue while reporting far smaller local – taxable – profits.

The ATO’s new ruling is an attempt to address part of that issue as it targets billions in untaxed offshore software royalties.

A broader interpretation

The new ruling broadens the circumstances in which software-related payments can be treated as royalties for tax purposes. Jethro Byrne, Grant Thornton corporate tax partner, says the ATO has formed the view that Australian distributors ‘effectively posses the rights to on-sell or distribute software, which in its view is generally a royalty’.

According to the ATO, payments may qualify as royalties when they are made for the use of, or right to use, copyright or similar intellectual property rights. The ruling points to situations where software intermediaries reproduce, communicate, modify or adapt software, or otherwise exercise rights associated with copyright ownership. It also addresses modern software distribution and SaaS arrangements.

The ATO has also made clear that it is particularly concerned about cross-border arrangements that may reduce or avoid tax on profits connected with Australia – an issue that has also long vexed many countries, including New Zealand.

For multinational software vendors, distributors and cloud providers, the Australian moves raise the prospect of much closer examination of how software revenues are structured and reported. Professional advisers have already warned them to reassess software licensing arrangements, treaty positions, withholding tax obligations and transfer pricing documentation in light of the ruling.

US pushback

Australia’s tougher stance hasn’t been universally welcomed and has put it at odds with some of its biggest trading partners.

Previous drafts of the ruling drew criticism from the US Treasury and the US National Foreign Trade Council business association argues the ruling ignores recent Australian case law and ‘further erodes investor confidence as well as the overall business climate in the country’.

One of the more contentious aspects of the ATO’s position is that the ruling applies to payments made both before and after its publications. The ATO maintains the ruling reflects its longstanding interpretation of the law rather than a policy change, a position that could expose some existing arrangements to review.

Critics also argue that Australia’s interpretation of software royalties is pushing beyond internationally accepted practice.

Extensions ahead?

The Australian Institute of Company Directors meanwhile, is warning that while the ATO’s ruling targets software and tech, ‘it’s implications could soon reach almost all Australian industries’.

“While end-user companies aren’t directly impacted, boards should be aware the new tax will likely be built into the future pricing of their software subscriptions,” it says, while also warning that other industries with value tied to intangible assets could be future targets. It suggests sectors including pharmaceuticals, cosmetics, motor vehicles and beyond could see similar attention in future.