AI adoption in Australia ranges from 12% to 72%. The numbers measure different things.

Ask how many Australian businesses use AI and the credible answers run from 12% to 72%. The Australian Bureau of Statistics says 12%. The National AI Centre puts it at 44%. The Reserve Bank's liaison survey says two thirds. auDA's small business survey says 72%. Every one of those figures is published, sourced and defensible, and a business trying to place itself against its peers has no way to choose between them.

Each is a defensible answer to a different question. Put each survey's base beside its headline and the spread resolves into a sequence: a market that has tried AI widely and embedded it narrowly.

That distinction carries more weight than any single number. What predicts whether AI changes a cost base is whether it sits inside a process that runs without someone watching.

The eight AI adoption figures, and what each one counts

Eight credible Australian sources, published between 2023 and 2026, put adoption somewhere between 12% and 72%. The base column is the part that matters.

Read down the base column and the ordering stops looking arbitrary. A narrow question asked of a broad population lands low, and a broad question asked of a selected population lands high. The ABS asks the whole business population whether AI appears on a list of technologies it uses, and gets 12%. auDA asks 408 small business owners whether they have used AI at all, and gets 72%. Both are telling the truth about what they measured.

Why the AI adoption surveys disagree

Four things separate these surveys, and only the first is usually acknowledged.

Start with the population. The ABS counts businesses, and Australian business is overwhelmingly small. Of the country's 2.73 million actively trading businesses, 1.74 million have no employees at all and another 689,000 employ between one and four people. The RBA's liaison panel is 105 firms chosen from the medium and large end. A rate measured across a population dominated by micro businesses cannot equal one measured across a panel of established firms. Neither is wrong.

The definition of use comes next. The ABS asks whether AI is among the technologies the business uses. The RBA asks how, and finds that nearly 40% of its adopters use AI for little more than summarising email in Copilot or ChatGPT. A single licensed seat counts as adoption under the first question and as minimal use under the second.

Depth is the third difference. The National AI Centre asks businesses to place themselves on a ladder from no adoption to full implementation. That produces a middle figure, 44% in February 2026, measuring the same ladder from a different angle.

Who answers is the fourth. Our own research asked 1,003 Australian adults about their own use of AI at work. Among workers under 45, 72% use AI in their role. Among workers over 45, 41% do. The RBA found the same pattern from the employer side and named it: adoption is often led by employees. The number rises when an employee answers, because most Australian AI use happens outside any formal adoption decision.

Weighting the ABS figures by employment puts adoption near 22%

Weight those rates by employment and the 12% and the two-thirds stop contradicting each other. The arithmetic uses the ABS's own published breakdown.

The ABS reports adoption by business size: 35% of large businesses, 22% of medium businesses, and about 11% of small and micro businesses. Those three rates describe different shares of the workforce. Small businesses with up to 19 employees hold 39% of private sector employment and medium businesses hold 26%. Large businesses hold 35%. Those shares come from the Small Business and Family Enterprise Ombudsman's calculations on ABS industry data.

Weight the adoption rates by those employment shares and the economy-wide figure lands at about 22%. The figure is an estimate: it assumes adoption is evenly spread inside each size band, but every input is a published number. The same survey that produces 12% produces roughly 22% once the unit of count is workers.

Both numbers answer a question. If you want to know how common formal adoption is across the business population, 12% is the answer. Count workers and the answer is closer to one in five: that is the share of the private sector workforce in a business where AI is in use.

The RBA's two-thirds describes the largest 7% of employing businesses

The gap between the RBA's 67% and the ABS's 22% for medium businesses is a sampling difference with a precise size.

Of Australia's 994,178 employing businesses, 67,857 employ between 20 and 199 people and 5,322 employ 200 or more. Together they are 7.4% of employing businesses. The RBA's panel of 105 firms is drawn from that group, and the Bank notes that participating firms "tend to be larger and established firms". The two-thirds therefore describes the top slice of the 7% of Australian businesses that employ 20 or more people, skewed further towards the largest within it.

That correction is the most useful one in this comparison. A 40-person professional services firm reading "two thirds of Australian firms have adopted AI" may conclude it is behind the market. It is comparing itself to a panel of banks, insurers and large employers. The relevant comparison for that firm is the ABS medium band, where the rate is 22%, and the National AI Centre's SME tracker, where it is 44%.

What the Reserve Bank found about depth of use

The Reserve Bank's November 2025 Bulletin is the most useful Australian source on what adoption looks like inside a firm, because it asked how deeply firms use AI.

Nearly 40% of firms reported minimal use, typically a digital assistant used for discrete tasks such as summarising email and research. Around 30% reported moderate use, with AI inside defined processes such as demand forecasting or inventory management. A small group had embedded AI across multiple business lines and used it in critical processes such as fraud detection. The Bank described the pattern as "piecemeal", and found adoption often "employee-led rather than employer-led".

The Bank also named why the productivity gains have not shown up yet. Firms told it that realising benefits depends on four things: hiring skilled people such as data engineers and product owners, redesigning workflows so AI sits inside decision-making, building data governance, and investing in retraining and change management. That is the complementary investment behind the productivity J-curve. The technology arrives before the workflow, the skills and the measurement, so the early returns are flat and the later ones are not.

Two further findings belong in any business case. Firms expect AI and machine learning investment to be much higher over the next three years than it has been. Around half expect AI adoption to reduce their total headcount slightly over that period. That expectation sits alongside the CSIRO's April 2026 finding. Firms which adopted AI between 2020 and 2023 advertised more jobs, not fewer, and at a faster rate than firms that did not.

What APRA found on AI governance

APRA has now published its own assessment of the regulated industries. The regulator wrote to every bank, insurer and superannuation trustee on 30 April 2026 after a deep-dive on a sample of the largest entities in late 2025. Its finding: AI adoption is accelerating while "governance, risk management, assurance and operational resilience practices are failing to keep pace with the scale, speed, and complexity of AI adoption".

The letter sets expectations against rules that already bind. CPS 230 on operational risk came into force on 1 July 2025, with amended text from 1 July 2026. CPS 234 on information security has required 72-hour incident notification since 2019. APRA expects boards to carry AI literacy and an AI strategy tied to risk appetite, and it flagged third-party and concentration risk, patching, and credible fallback processes for critical operations. ASIC reached a similar conclusion in Report 798. The two regulators ran joint frontier AI roundtables in June and July 2026, then urged entities on 27 August 2026 to move from awareness to action.

The ABS, the RBA and APRA reach the same finding from three directions. Adoption has run ahead of the operating discipline that would make it pay.

The AI6 checklist the government published

The National AI Centre published the Guidance for AI Adoption on 21 October 2025. It condenses the ten guardrails of the earlier Voluntary AI Safety Standard into six practices, which the market now calls AI6:

1. Decide who is accountable.

2. Understand impacts and plan accordingly.

3. Measure and manage risks.

4. Share essential information.

5. Test and monitor.

6. Maintain human control.

The guidance is voluntary. The National AI Plan of December 2025 confirmed Australia will rely on existing law plus AI6, with no standalone AI Act. Voluntary in the way the Essential Eight is voluntary: it is becoming the benchmark that boards, auditors and large customers ask about. It ships in two versions, foundations for organisations using AI in low-risk ways and implementation guidance for teams building AI or running higher-risk systems. The implementation version separates actions for developers, deployers and general-purpose AI providers.

For a business working out what to do next, AI6 is the most useful artefact the government has published, because it gives you six things to check. The gap it exposes is operational. Most organisations can name the tools their people use. Fewer can name who is accountable for a given AI use case, what its impacts were assessed to be, or how it is monitored once it is live.

How organisations close the AI adoption gap

Organisations close the gap by taking a process, putting AI inside it, and keeping a person accountable for the outcome.

Our own research suggests the appetite is already there. In a survey of 1,003 Australian adults, people who use AI at work reported saving two hours a day, or 520 hours a year. Only 17% used that time to work fewer hours. 83% put it back into work, and 58% reinvested it specifically into learning and process improvement, which is the behaviour that compounds. The constraint on turning that into a business result is whether the organisation has redesigned the process around the tool, which is the same conclusion the RBA reached from the employer side.

That work is unglamorous and specific. The voice triage work we did with healthdirect turned on what happens when the model is wrong, who takes over, and how the handover is measured. The real-time voice prototype we built and wrote up had the same shape: most of the engineering sat in latency, interruption handling and failure paths. The model was the easy part.

Common questions about AI adoption in Australia

What percentage of Australian businesses use AI?

12% of businesses reported using AI in 2024-25, according to the ABS Business Characteristics Survey. Weighted by employment, the same data implies roughly 22% of private sector workers are in a business that uses AI. The higher figures in circulation, up to 72%, come from surveys with narrower populations and broader definitions of use.

Why do the surveys disagree so much?

They measure different populations, define use differently, and ask at different depths. The ABS asks every kind of business whether AI is on its technology list. The RBA asks 105 large firms how deeply they use it. The National AI Centre's tracker asks SMEs to rate their own stage of adoption. Each is a sound measurement of its own question.

Is AI adoption in Australia low compared with other countries?

The RBA found Australia at the low end of advanced economies on both adoption and trust. It also noted that Australia invests relatively little in ICT and software as a share of total investment. The ABS figures show the gap closing quickly: large business adoption rose from 9% to 35% in three years, and financial services rose from 1% to 24%.

What is the AI6?

The six essential practices in the National AI Centre's Guidance for AI Adoption, published 21 October 2025: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, and maintain human control. It is voluntary and is becoming the de facto benchmark for AI governance in Australia.

How many Australian employees use AI at work?

72% of workers under 45 and 41% of workers over 45 report using AI in their role, from our survey of 1,003 Australian adults. Among workers aged 18 to 24 the figure is 80%. That is well above the share of businesses reporting formal adoption, which is the employee-led pattern the RBA describes.

Where these AI adoption figures come from

Every figure in this article comes from a published source, dated, with its survey base attached. The employment-weighted estimate of 22% is arithmetic over the ABS size-band rates and the Ombudsman's employment shares, and it carries the estimate label because the within-band distribution is assumed. The chart accompanying this article plots all eight business-level figures against one axis and marks that estimate.

Two consultancies published shorter versions of this observation in 2026. What this article adds is the arithmetic. The employment weighting closes the gap between 12% and 22%, and the firm-count calculation puts the RBA's panel inside the 7.4% of employing businesses with 20 or more staff.

If your organisation wants to know where it sits against these numbers, ask a narrower question. Which of your processes has AI inside it, who is accountable for that process, and how would you know if the AI stopped working?

If your organisation is using AI and you can't answer those three questions, get in touch.