Executive summary
Key takeaways for business leaders
Australia’s productivity problem is real and worsening. Labour productivity in mid-2025 sat at 2019 levels. The Reserve Bank of Australia now assumes just 0.7% annual growth, against 2% in the US and 3%+ in South Korea. Without a step change in how we work, living standards plateau.
AI is the most consequential lever available to business leaders right now. But, most organisations aren’t pulling it effectively.
- 87% of enterprises are using AI somewhere. 95% are seeing zero return on their generative AI investments
- Nearly two-thirds remain stuck in pilot purgatory, experimenting without scaling
- The failures are almost never technical. They are organisational: vague problem definitions, poor data readiness, unchanged workflows, and absent governance
- Australia has more to gain than most. A service-heavy economy, scarce and expensive labour, and a wider productivity gap mean the returns on effective AI adoption are proportionally higher here
- The risk is equally asymmetric. Over 60% of Australians report low AI knowledge. Only 24% have received any AI training. We are behind Singapore, China, and South Korea on capability, and falling further back
- A market correction is plausible. Analysts assign roughly 45% probability to an extended reset. Organisations without measurable returns will have nothing to defend when budgets tighten
- The organisations seeing real results share one trait: they started with a specific, measurable problem and executed with discipline
- Workforce transition is not optional. Up to 1.3 million Australians may need to move into new roles by 2030. Organisations that ignore this will face a backlash that outlasts the gains
The bottom line: Australia doesn’t have a technology problem. It has an implementation problem. The organisations that treat AI adoption as a discipline challenge rather than a technology bet are the ones that will still be growing when the hype cycle ends.
By Jonathan Hardy | Principal Consultant – AI & ML, Mantel
If you look at the numbers, Australia’s labour market looks healthy: near-record employment, with 14.7 million people in work. Unemployment is around 4%, and skilled migration is returning to pre-pandemic levels, with 185,000 permanent places allocated for 2025-26. But the number that matters most tells a different story: labour productivity. In mid-2025, labour productivity was essentially back at 2019 levels. In 6 years, the economy has added people and hours… but it hasn’t added output per hour. This represents five years of zero net progress.
The Productivity Commission’s 2025 bulletin makes for uncomfortable reading. Multifactor productivity declined 0.5% over 2024-25. The Reserve Bank of Australia has downgraded its medium-term labour productivity assumption to just 0.7% per annum. The US manages closer to 2%; innovation-driven economies like South Korea run at 3% or higher. A 1% difference, sustained over a decade, becomes a 10% cumulative divergence: in real wages, in competitiveness, and in what Australians can actually expect from their standard of living.
Weaker productivity constrains the economy’s potential output. Treasury projections show GDP growth is expected to strengthen, but the productivity drag limits how much of that translates into real income gains. When productivity stalls, the only way to grow the economy is to add more people and more hours, and we’re already near the ceiling on both. The implication is stark: without a step change in how we work, productivity and living standards will plateau.
To be clear, productivity is not a single-variable problem. The Productivity Commission’s own analysis identifies regulatory burden, weak competitive dynamism, and decades of underinvestment in enabling infrastructure as structural headwinds. AI won’t resolve any of those. It won’t rewrite planning laws or streamline the tax code. But among the levers business leaders can actually reach, I think effective AI adoption presents a genuine opportunity to tackle Australia’s stalling productivity growth.
Defining productivity in the Australian context
Productivity measures how efficiently inputs (such as capital and labour) are used to produce outputs (goods and services). Productivity growth occurs when an economy produces more output with the same input, maintains output levels using fewer inputs, or a combination of both (source: APH).
Closing the AI productivity gap
AI holds extraordinary potential, yet a significant gap exists between its promises and the actual value most organisations are deriving from it. Enterprise adoption has reached 87% at the “using AI somewhere” level. Global spending has grown from $1.7 billion to $37 billion since 2023. The MIT Media Lab reported in August 2025 that 95% of organisations are seeing zero return on their generative AI investments. A February 2026 NBER study found 90% of firms reporting no productivity impact. Nearly two-thirds remain in “pilot purgatory”, running experiments, declaring success, and never actually scaling.
At Mantel, we see this constantly, and it follows a familiar pattern. Ambitious internal announcements. Real budget commitments. A handful of pilots that show promise. And then, somewhere between the pilot and the rollout, the returns quietly fail to materialise.
What strikes me, when I look at stalled AI initiatives up close, is where the problem lies. The technology works; of that, there is no doubt. Instead, the failures are organisational. Vague problem definitions where “improve productivity” substitutes for a specific, measurable objective. Data scattered across legacy systems that AI can’t effectively reach. No definition of what “good” looks like before deployment. Governance paralysis, genuine uncertainty about data privacy, model risk, and responsible use that jeopardises decision-making. And tools deployed into workflows that stay fundamentally unchanged around them, so the efficiency gain never materialises.
“The organisations seeing real returns share a common trait: they execute AI initiatives with discipline. They identified a problem specific enough to measure, applied AI against it with rigour, and looked honestly at the results. That's how you move from superficial adoption to productive adoption, with demonstrable ROI that builds the confidence to invest further.”
Jonathan HardyPrincipal Consultant - AI & ML, Mantel
Superficial adoption and deep, productive adoption look similar from the outside. The investment is real in both cases. The difference shows up in the results, and right now the results are telling us something most organisations don’t want to hear.
Comparison Table: Superficial vs Productive AI Adoption
| Feature | Superficial Adoption | Productive AI Adoption |
| Objective | Abstract mandate to “be AI-enabled” | Specific, measurable business problem |
| Workflow | Tools added to unchanged processes | Workflows redesigned to capture value |
| Measurement | No clear definition of success | Disciplined ROI tracking and rigour |
| Workforce | Low training and limited engagement | Heavy investment in reskilling and readiness |
The dot-com parallel – Avoiding an AI investment correction
Many economists and technologists have drawn parallels between the AI boom and the dot-com era of the early 2000s. Market concentration in the five largest companies hasn’t been this high since the late 1990s. Bridgewater’s Ray Dalio has described current AI investment levels as “very similar” to dot-com. Tech giants are on track to pour $350-500 billion into AI infrastructure in a single year, with cumulative spending projected to hit $3 trillion by 2028.
The parallel isn’t about the technology failing. The internet didn’t fail. It’s about the gap between investment and returns becoming unsustainable if organisations don’t move past experimentation. Analysts are currently assigning roughly a 45% probability to an “extended correction” scenario; a rolling reset where the weakest players will fall first.
For Australian businesses, a correction wouldn’t be a Silicon Valley problem watched from a distance. We’re a smaller addressable market, which means global AI vendors would scale back here faster than elsewhere. AI specialists would follow the funding to larger markets. And inside organisations, a failed investment cycle leaves something harder to fix than a budget gap: a credibility deficit for AI as a productivity tool that makes the next initiative harder to sponsor even when the underlying technology has continued to improve.
This is precisely why disciplined, ROI-focused adoption matters. Organisations that can point to proven, measurable returns from their AI investments will handle any correction with their momentum intact. Those running expensive experiments without clear business cases will be the first to falter.
The central challenge remains: can Australia convert meaningful AI adoption into measurable, sector-wide productivity gains?
Why Australia has more to gain, and more to lose
There’s a case that Australia is better placed than most to benefit from AI-driven productivity gains. Not because we’re ahead on adoption, but because our structural conditions make the returns proportionally higher. Labour is scarce and expensive. The workforce is ageing. The economy skews heavily toward services. When AI can augment an expensive worker in a tight labour market, the value of that augmentation compounds in ways it simply doesn’t in lower-cost economies. And with our productivity gap wider than most advanced economies, the ceiling for improvement is higher too.
The forecasts tell a similar story. OpenAI’s economic blueprint puts AI’s potential contribution to Australia at A$115 billion annually by 2030, with A$80 billion of that driven by productivity gains. The Productivity Commission estimates $21 billion in annual benefit already available, with sevenfold growth possible by the end of the decade.
However, productivity gains and workforce disruption are two sides of the same coin. If AI delivers on these forecasts, the composition of work will shift materially.
McKinsey estimates up to 1.3 million Australian workers may need to transition into new roles by 2030. Administration and financial processing are already being restructured. CBA’s $90 million workforce transition programme is an early and visible signal of a shift that is happening across sectors, mostly quietly. The organisations treating AI purely as a growth story, without reskilling investment and honest communication, will find that the productivity gains come with a political and human backlash against AI that’s hard to unwind once it starts.
What makes this harder is that Australia is starting from behind on the capability side. Over 60% of Australians report low AI knowledge, against 48% globally. Only 24% have received any AI training, compared to 64% in China and 45% in Singapore. Our AI wage premium sits 6% below the global average. The countries moving fastest on AI skills and infrastructure aren’t waiting for the right moment: they’ve already started.
In short, Australia has a structural advantage in needing AI more than most, but at the same time we risk squandering it by being slower to build the capability to use it well.
Moving from AI experimentation to measurable outcomes
So how do organisations actually move from potential to results? AI lifts productivity through three distinct channels. The first is task automation, doing existing work faster and at lower cost. The second is augmentation, enabling workers to produce higher-quality output than they could alone. The third is innovation, creating entirely new services, products, or business models. For Australia’s service-heavy economy, the augmentation channel may matter most: AI doesn’t replace the financial analyst, the case worker, or the project manager. It makes each of them materially more productive per hour worked.
We work with organisations across enterprise, government, and sport who’ve moved from “let’s explore AI” to “AI is now core to how we operate.” The ones creating real value and those stuck in endless experimentation are separated, almost always, by the same thing: their internal grasp of what AI can actually do, combined with the discipline to pursue it in a focused way rather than a sweeping one.
In practice, this means starting with a specific, measurable problem rather than an abstract mandate to “be more AI-enabled.” It means working closely with the people who actually do the work to understand where AI fits into their existing workflows, or where it might enable entirely new ones. It means investing in how people think about their work, not just giving them access to a tool and hoping they figure it out.
Deploying Claude or ChatGPT takes five minutes. Redesigning a workflow to actually capture the value can take months. Building the governance framework that makes safe, scalable deployment possible takes more effort still. Most organisations underinvest in the last two and then wonder why the first one didn’t change anything.
There’s also a psychological dimension that gets underestimated. When you set realistic expectations about what a given AI deployment will achieve, what it will cost, how long it will take, what success actually looks like, you create the conditions for genuine surprise. A claims processing team that expected a 20% time saving and got 40% is energised. They trust the next AI initiative. They become advocates for change.
What this means for Australian organisations
The organisations we see getting stuck are almost always the ones that began with a sweeping vision rather than a specific problem. Identify a genuine opportunity – a process that’s too slow, a decision that’s too manual, a team that’s stretched too thin, and apply AI to that problem with a committed team and a disciplined approach to measuring what happens.
Be honest about what implementation actually takes. Only a quarter of Australians have received any AI training. That’s both a problem and an opportunity: the upside of investing in your people is enormous precisely because so few organisations are doing it well. The organisations that invest in workforce readiness see materially better returns.
And be honest about the transition. AI will change the composition of work in your organisation. Some roles will be augmented, some will be restructured, and some will eventually disappear. Pretending otherwise isn’t optimism. It’s the kind of leadership that erodes trust and may produce a backlash that makes the next AI initiative harder to land. The organisations that navigate this well will be the ones that communicate openly, invest in reskilling, and treat workforce transition as a first-order priority rather than something to address later, when later turns out to be too late.
The discipline dividend
Australia doesn’t have a technology problem. We have an implementation problem, compounded by structural headwinds that technology alone cannot overcome. Regulatory complexity, weak competitive dynamism, and decades of underinvestment in enabling infrastructure all weigh on the productivity numbers. AI won’t fix planning laws or simplify the tax code. But among the levers that business leaders can actually pull, it may be the most consequential in a generation.
The dot-com parallel is instructive here. The internet didn’t fail when the bubble burst – it just got serious. The companies that had built real, measurable value survived the correction and emerged as the most consequential businesses of the next two decades. If a similar reset comes for AI, the organisations left exposed won’t be the ones that invested in AI. They’ll be the ones that invested in AI hype: big commitments, unclear returns, no measurable business case to defend when budgets tighten.
The question for Australian businesses isn’t whether to adopt AI. Most already have. The question is whether we’ll move past the experimentation phase with enough discipline to capture the productivity gains that are genuinely available, through automation, augmentation, and entirely new ways of working, or whether we’ll let the hype cycle run its course and look back in five years wondering why the needle still hasn’t moved.
The last five years of stalled productivity growth is what got us here. The next five don’t have to look the same. But that’s not a technology bet, it’s a discipline bet. Underpromise on what AI will deliver next quarter. Overdeliver on what you actually build. That’s how you turn a productivity challenge into a productivity advantage.
Why is Australia’s productivity growth stalling?
Australia’s labour productivity has returned to 2019 levels, representing five years of zero net progress. This stagnation is driven by structural headwinds like regulatory burdens and underinvestment in infrastructure, alongside a failure to convert new technologies into measurable output gains.
How can AI improve business productivity?
AI enhances productivity through three primary channels: task automation, human augmentation, and business model innovation. In Australia’s service-oriented economy, augmentation provides the most value by allowing professional workers to produce higher-quality output in less time.
Why are most companies failing to see a return on AI?
Failure usually stems from organisational issues rather than technical ones. Common barriers include vague objectives, poor data quality, and deploying tools into outdated workflows without changing how the work is actually performed.
What is the “AI capability gap” in Australia?
Over 60% of Australians report low AI knowledge, and only 24% have received formal AI training. This puts Australia behind regional competitors like Singapore and South Korea, where training rates and AI wage premiums are significantly higher.
How many Australian jobs will be affected by AI?
Estimates suggest that up to 1.3 million Australian workers may need to transition into new roles by 2030. While many roles will be augmented, others in administration and financial processing are already undergoing significant restructuring.
