On September 3, 2026, Treasurer Jim Chalmers called artificial intelligence the “biggest economic transformation in our lifetime,” but a new Treasury note suggests the payoff is far from guaranteed. The paper, reported by Information Age (ACS), says about two‑thirds of Australian companies use AI in some form, yet fewer than one in ten call that use “significant.” That mismatch matters for Australia AI productivity: Treasury estimates AI could lift productivity by 1.5–2% a year, but it kept the official forecast pinned at 1.2% because adoption remains shallow and uneven.
What Treasury’s numbers really say about Australia AI productivity
The gap between potential and forecast is the story. Treasury’s estimate of a 1.5–2% annual boost is the kind of lift policymakers dream about in a country where productivity growth has struggled for years. The Productivity Commission’s five‑year inquiry underscored that drag and argued that diffusion of proven technology is the main lever left. Chalmers’ warning lines up with that logic: if most firms only dabble with pilots, the aggregate needle will barely move.
The use‑case split explains why. Admin copilots and point solutions trim small tasks but seldom touch the core process metrics that drive Australia AI productivity. “Significant” use, as implied by the Treasury note cited by Information Age, looks more like re‑platforming customer support with retrieval‑augmented systems, algorithmic scheduling tied into ERP, or AI‑assisted code generation rolled out across engineering teams with measurable throughput gains. The first bucket saves minutes; the second rewrites cost curves.
Treasury’s two‑speed economy warning follows from that. If the firms that go deep compound faster, and the rest keep trials on the shelf, the gap widens even if headline AI usage looks high. That’s how you get a high‑adoption, low‑impact paradox.
Where AI productivity in Australia will land first
Early gains usually accrue to companies with clean data, process discipline, and capital to absorb the upfront spend. That’s not conjecture. OECD research has found AI diffusion tends to concentrate among larger, digitally mature firms, with weaker uptake among smaller businesses. Translate that to the Treasury numbers and you get a map: big banks, miners, retail networks, and healthcare groups likely bank the first wave; small and mid‑sized firms risk falling a step behind.
The consequence for Australia AI productivity is skew. When a few large players scale AI into core workflows, sector averages rise modestly while long tails of shallow adoption pull the aggregate back toward that 1.2% forecast. That makes the mix of adoption—not just the rate—decisive.
There’s also a workforce angle. If AI depth clusters in certain firms, so does demand for skills: data engineers, ML platform leads, prompt and evaluation specialists, and product managers who can thread AI into revenue lines. Without a broader upskilling push, labor scarcity becomes a ceiling on diffusion, which keeps national productivity below its AI‑assisted potential.
Policy moves Canberra can make after the Treasury AI analysis
The Information Age report says governments are working through how to protect Australians while embracing the upside. The balance matters because risk rules set the speed limit for adoption. Canberra’s consultation on Safe and Responsible AI in Australia has been running since 2023; the next step is aligning guardrails with diffusion, so risk‑sensitive sectors can move beyond experiments without guesswork.
Four levers would help close the gap Treasury identified and lift Australia AI productivity toward that 1.5–2% range:
- Clarity, then consistency: Finalise targeted, risk‑based rules and testing regimes so regulated sectors can scale deployments without months of legal dead time.
- Skills where the bottlenecks are: Fund short, stackable credentials in data engineering, model evaluation, and secure ML ops. Tie support to evidence of workplace deployment, not just enrolments.
- Procurement as a market‑maker: Require measurable AI outcomes in public‑sector projects—latency cutoffs, error budgets, and productivity KPIs—to create referenceable wins SMEs can point to.
- Incentives that reward depth: Tilt investment allowances toward projects that integrate AI into core systems (ERP, CRM, EMR), with reporting on realised productivity, not just spend.
None of this reduces the need for standard consumer protections. It aligns them with adoption so firms don’t stall on the threshold. The prize—higher and broader Australia AI productivity—depends on it.
What leaders should do now to capture Australia AI productivity gains
Waiting for perfect certainty has a cost. The Treasury forecast itself bakes in today’s hesitation. Leaders can change that trajectory by moving from pilots to production with a bias for measurable outcomes:
- Pick one end‑to‑end workflow and instrument it. Define a baseline, deploy an AI‑assisted redesign, and track throughput, errors, and cycle time every week for a quarter.
- Fix data before features. Consolidate the minimum viable dataset, set retention and access rules, and implement human‑in‑the‑loop checks where model outputs touch customers.
- Build an evaluation habit. Agree on a small set of task‑level benchmarks and red‑team prompts; review them in the same cadence as financial KPIs.
- Train the team that touches the work. Pair power users with engineers; publish playbooks; reward time saved or revenue created, not just bot launches.
- Mind security and provenance. Adopt content credentials where practical and keep a short list of approved models and plugins with auditable logs.
These steps don’t require a moonshot budget. They do require picking spots where AI can move the core and proving it with numbers. That’s how firms turn “use” into “significant use,” and how the aggregate moves from 1.2% toward Treasury’s upper bound.
Why the timing matters
Chalmers’ assessment—that AI will touch every part of the economy—matches the scale of the opportunity and the risk. The Treasury note frames both: widespread dabbling without depth risks a two‑speed economy; deliberate scaling can pull national output up. Australia AI productivity won’t rise because tools exist. It will rise if firms and governments turn them into redesigned work, backed by rules and skills that make scaling safe to do.
The window is open. The forecast says it won’t stay that way by itself. For more on this, see bloomberg.com and nytimes.com.