How many Australian businesses actually use AI?
Ask four reputable Australian sources how many businesses use AI and you can get 12%, 41%, two thirds or 72%. The figures are not competing estimates of the same thing. The surveys ask different businesses different questions at different points between trying an AI assistant and embedding AI in a working process. This brief puts the sample, definition and adoption depth beside each number so an owner can see what the data supports.
Josh, published 2 September 2026
The numbers disagree only when the labels disappear
The evidence points to common experimentation. It does not show that deeply embedded, measured use is common across Australian business.
The Australian Bureau of Statistics found 12% of employing businesses used AI during the 2024 to 2025 financial year. The ABS also states that its question was not designed to measure the intensity or extent of that use.
auDA found 72% of 408 surveyed small businesses had used AI in 2026. The National AI Centre reported 41% of small and medium enterprises were adopting AI in June 2025. The RBA found two thirds of 105 mostly medium and large liaison firms had adopted it in some form.
None of those sources is necessarily wrong. The ABS covers the employing business population with exclusions and asks about business technology use. auDA asks a smaller sample of owners and managers about any AI use. The National AI Centre tracks stages of adoption among businesses with up to 500 employees. The RBA studies a selected group of established firms and asks how deeply the technology has been put to work.
One market, four questions
Primary Australian sourcesThe defensible conclusion
Finding- AI experimentation is widespread. Broad surveys pick up staff and owners trying assistants for ordinary tasks.
- The survey frame changes the rate. A whole business technology question can produce a smaller figure than a broad question about any use.
- Deep adoption is smaller again. The RBA found much of the use in its relatively advanced sample was still minimal.
Keep the method beside the number
The source ledger turns four incompatible headlines into four useful measurements.
| Source | Figure | Survey base | What counted | Limit |
|---|---|---|---|---|
| ABS | 12% | Employing businesses, with stated exclusions | AI selected from a list of business technologies used | The question did not measure intensity or extent |
| National AI Centre | 41% | 400 small and medium enterprise responses each month, weighted by industry, location and employee count | Businesses described broad use, limited use or implementation through staged response options | The June 2025 release uses the broader phrase adopting AI |
| RBA | Two thirds | 105 liaison firms, mostly medium and large | AI adopted in some form, with depth asked separately | The sample is weighted towards established firms and does not represent small business |
| auDA | 72% | 408 small business owners or managers in an online survey | Respondents reported having used AI | Any use can include experimentation by one person |
How to quote the data
Source ruleTrying a tool is not the same as changing a process
A business can be using AI while having no approved tool, no process owner and no measured result.
The National AI Centre survey shows why a single yes or no question is too blunt. It separates broad use, limited use, implementation planned within 12 months, no intention to implement and no awareness of how AI could be used. The tracker also asks which applications are in use and what responsible practices sit around them.
Jobs and Skills Australia uses a capability spectrum that moves from adoption to integration and maturity. It also treats staff use without formal approval as an early signal, not evidence that the organisation has integrated AI properly.
For a small business, the useful distinction is operational. Someone opening ChatGPT for a draft is use. A licensed tool with approved data rules is sanctioned use. AI built into a recurring workflow is embedded use. A workflow with a baseline and a reviewed outcome is measured use.
A practical adoption scale
Working framework- Tried. A staff member or owner has used an AI tool for a task.
- Sanctioned. The business has approved the tool, data boundary and accountable owner.
- Embedded. AI is part of a recurring workflow with review and a failure path.
- Measured. The business compares a defined result against a baseline.
Most business use is still close to the surface
Nearly 40% of the RBA sample reported only minimal AI use.
The RBA's 2025 liaison survey is useful because it asked about depth after establishing whether a firm had adopted AI. Two thirds had adopted it in some form, but nearly 40% of the full sample reported minimal use. Those firms commonly used assistants such as ChatGPT or Microsoft Copilot for isolated work including research and email summaries.
Around 30% of the full sample had made more substantive progress. Moderate adopters used AI inside processes such as forecasting or inventory management. A smaller group had integrated it across several business lines or into critical work such as fraud detection.
The sample was already weighted towards larger and more established firms. Shallow use inside that group is a warning against treating an account licence or an isolated prompt as proof of business change.
Use is not yet value
Outcome check- 22% reported faster decisions. The June 2025 National AI Centre release records this as a business benefit.
- 18% reported better productivity. This is a reported outcome, not a measured economy wide productivity effect.
- 7% measured digital contribution. The ABS figure covers all digital activity, not AI alone, but shows how uncommon business measurement remains.
Use inside the RBA sample
105 liaison firms, 2025The national average hides a wide business gap
ABS industry estimates range from 1% to 38%.
In the ABS data, information, media and telecommunications recorded 38% AI use. Professional, scientific and technical services and financial and insurance services each recorded 24%. Construction recorded 6%, accommodation and food services 5%, agriculture 3%, and transport, postal and warehousing 1%.
Innovation status widened the gap again. Among small businesses with 5 to 19 employees, 19% of innovation active firms reported AI use compared with 4% of firms that reported no innovation activity.
Jobs and Skills Australia describes adoption as moving at different speeds. Large firms and market sector organisations are further ahead, while investment capacity, digital foundations, leadership, data, governance and workforce skills influence whether a tool moves beyond experimentation.
AI use by selected industry
ABS, 2024 to 2025What an average misses
Context- Industry changes the available tasks. Office based information work has more obvious entry points than physical work.
- Capability changes adoption depth. Data, systems, management time and staff skills determine whether an experiment becomes a process.
- Survey samples change the headline. A sample of established liaison firms will not read like the wider employing business population.
Count the work, not the licences
Your useful adoption rate is the share of important workflows that are owned, controlled and measured.
Start by recording the AI tools staff already use, including personal accounts. The free Australian AI register template gives the inventory a practical home. Record the tool, task, information entered, owner and current approval status.
Sort each use into tried, sanctioned, embedded or measured. Do not push every experiment towards integration. A task that handles sensitive data, affects a customer or creates a legal decision needs a stronger boundary than a draft internal summary. The AI policy template covers those rules.
For one repeatable workflow, record the current time, error rate, rework or response time. Run the changed process for a fixed period and review the same measure. A reported productivity benefit is useful. A result compared with a baseline is better.
The Productivity Commission's estimate of a 116 billion dollar economic opportunity over a decade is modelled potential. It is not value that Australian businesses have already banked. The gap between potential and performance is where process design, staff training and management judgement sit.
A 30 day adoption check
Practical action- Inventory current use. Include personal tools and informal workarounds.
- Mark the adoption stage. Separate experiments from approved and embedded work.
- Set one baseline. Measure a useful workflow before changing it.
- Review the evidence. Keep, change or stop the use after 30 days.
Frequently asked questions
The ABS found that 12% of Australian employing businesses used AI during the 2024 to 2025 financial year. Other reputable surveys report 41% of small and medium enterprises adopting AI, two thirds of mostly medium and large firms using it in some form, and 72% of surveyed small businesses having used it. The figures answer different questions, so there is no single percentage that describes every kind of business use.
Every figure with its boundary
Related Field Notes
Read nextFind the adoption you already have