88% of Australian businesses use AI. Only 10% embed it
Everyone has a ChatGPT login; almost nobody has a process that runs on it. Pick one process with a measurable cost, connect it to your system of record, check the step that matters and measure it for 30 days.
On 9 October 2026 CPA Australia released its Business Technology Report 2026. It found 88% of Australian businesses use AI, but only 10% use it widely and have deeply embedded it. Of the markets the report compares, Australian businesses were the least likely to say AI runs across several parts of the business.
The gap between those two numbers is the difference between trying AI and running a process on it. Everyone has a ChatGPT login. Almost nobody has a process that runs on it, with a named owner, a connection to the system where the work actually lives, and a number that shows whether it is paying its way.
For a 50 to 250 person business, closing the gap does not start with a bigger AI budget. Pick one process with a cost you can measure, write down its steps, connect it to your system of record, put a person on the step that matters, and measure it for 30 days. Then do the next one.
What did CPA Australia actually find?
The survey ran in July and August 2026 and collected 1,679 responses from accounting and finance professionals in Australia, the Chinese Mainland, Hong Kong, Malaysia and Singapore. So these are finance people describing their own organisations, which is worth keeping in mind. Here are the Australian numbers, from the report and CPA Australia's Australian infographic:
| Measure | Australia | Comparison |
|---|---|---|
| Use AI in some form | 88% | 89% across all markets |
| AI used widely and deeply embedded | 10% | 14% across all markets |
| AI used in several areas, not deeply embedded | 43% | 42% across all markets |
| Use agentic AI | 28% | 58% in the Chinese Mainland |
| Use custom-built or in-house AI tools | 36% | 61% in the Chinese Mainland |
| Staff use AI with little or no oversight or guidance | 26% | 29% of organisations across all markets |
Two more findings matter for a mid-sized business. Organisations under 500 staff were less likely to have AI deeply embedded (12%) than those with 500 or more (18%). And 63% of businesses that grew their profitability said AI was used across several functions or the whole organisation, against 50% of those whose profitability stayed flat or shrank. The report is careful to say that is a link, and it does not prove cause.
CPA Australia's Business and Investment Lead, Gavan Ord, put it plainly in the release: AI adoption alone isn't enough.
Share of businesses using each. Source: CPA Australia, Business Technology Report 2026 and its Australia infographic.
Is this the same story as the August numbers?
Yes, with a sharper edge. In August we pulled together what the numbers actually say about AI adoption in the Australian mid-market, and the finding was that surveys disagree because they measure different things: one person using a chatbot, or a business running a system. Treat this post as the October update.
CPA's survey asks the second question directly, and it gives a clean split: 88% use, 10% embedded. The National AI Centre's tracker, which surveys around 400 SMEs a month, puts the share of SMEs using AI lower, at 43% for December 2025 to February 2026, the most recent quarterly overview we could find at the time of writing. It also reported that broad adoption, AI used across several parts of the business, was at its highest level in seven months. Different samples, same direction: the people already using AI are going deeper, and most businesses are still at the login stage.
How do you tell if your business has tried AI or embedded it?
Ask where the work happens. If AI only appears when someone remembers to open a chat window, you have tried it. If a piece of work moves through AI as a normal step, every time, and someone would notice if it stopped, you have embedded it. Here are the signs we look for:
| Tried it | Embedded it |
|---|---|
| Staff have personal ChatGPT or Copilot logins and use them when they think of it | A named process uses AI as a fixed step, whether or not anyone remembers |
| Prompts live in people's heads or a shared doc nobody updates | The instructions are written down, versioned and owned by one person |
| People copy and paste between the AI and HubSpot, Xero or the ERP | The AI reads from and writes to the system of record directly |
| Nobody can say what it saved last month | There is a before and after number: hours, turnaround time, error rate |
| Checking the output is up to whoever used it | A specific person checks a specific step, and the check is logged |
| If the keen user leaves, the AI use leaves with them | It keeps running when people change roles |
| The AI policy, if there is one, is a list of banned tools | The policy says who approves a new use and who signs off on each process |
If most of your ticks are in the left column, you are in the 78% between using and embedding, and you are in big company. The useful part is that one process is enough to move a business into the right-hand column.
Australian businesses. Source: CPA Australia, Business Technology Report 2026, released 9 October 2026.
Which process in your business should run on AI first?
We can work through your shortlist with you and pick the one with a cost you can measure.
How does a 50 to 250 person business close the gap?
One process at a time, in this order. Each step is small. Skipping one is how pilots stall.
- 1. Pick one process with a cost you can measure. Good first candidates are repetitive, high volume and already measured, or easy to measure: triaging inbound enquiries, drafting first-pass quotes, coding supplier invoices, writing call summaries into the CRM, answering the standard sections of a tender. Write down what it costs today in hours a week or days of turnaround. If you cannot put a number on it, pick a different process.
- 2. Write down the steps as they actually happen. Sit with the person who does the work and list every step, including the workarounds. Most of the value comes from this list, because it shows which steps need judgement and which are copying, sorting and summarising.
- 3. Connect it to the system of record. If the work lives in HubSpot, the AI reads from and writes back to HubSpot. Copy and paste is the single biggest reason a trial never becomes a process: it depends on someone choosing to do it, every time.
- 4. Put a person on the step that matters. Pick the point where a mistake costs money or trust, such as a price going to a customer or a claim in a tender, and make a named person approve it. Everything before that step can run on its own.
- 5. Measure for 30 days. Compare against the number from step one. Track how often the person at the check changed the output, because that tells you whether the instructions need work.
- 6. Then the next one. The second process goes faster, because the connection, the policy and the habit of measuring already exist.
A typical first process takes a few weeks to set up properly. The time goes into steps two and three, and very little into the AI itself.
What usually goes wrong?
From what we see on discovery calls, the same four things stall most attempts:
- Starting with the tool. Buying licences for everyone and waiting for use cases to appear. You get 88% usage and nothing embedded, which is exactly the national picture.
- Picking a process nobody measures. If there is no before number, there is no after number, and the project loses its funding at the next budget review.
- No owner. An AI process needs a person who owns the instructions, the check and the number. Without one it drifts until it stops.
- No rules for staff. CPA found 26% of Australian organisations let staff use AI with little or no oversight or guidance. That is where customer data ends up pasted into a personal account. A one-page policy fixes most of it; our post on who signs off on AI has a version you can copy.
What should you do this week?
- List five candidate processes and the rough weekly hours each one takes. Pick the one with the clearest number.
- Name an owner for that process and give them a few hours a week to work on it.
- Check where its data lives. If the answer is HubSpot or your ERP, find out whether your AI tool can connect to it properly before you build anything.
- Write or update your AI policy so it covers approved tools and who signs off on new uses.
If you are planning a wider rollout once the first process works, our guide to rolling Claude out to a 100-person business covers the training, the access and the order we do it in.
Frequently asked questions
What percentage of Australian businesses use AI in 2026? CPA Australia's Business Technology Report 2026, released on 9 October 2026, found 88% of Australian businesses use AI in some form. Only 10% said AI is used widely and deeply embedded across the business.
What does it mean to embed AI in a business? It means AI is a fixed step in a named process, connected to the system where the work lives, with an owner, a human check where mistakes are costly and a number that shows the result. Occasional use of a chatbot by individual staff is trying AI.
Why is Australia behind on AI integration? CPA Australia's report does not give a single cause. It found Australian businesses were the least likely of the markets compared to use AI across several parts of the business, and it lists privacy, security, overreliance on AI and reduced human oversight as the main barriers across all markets.
What is the first AI process a mid-sized business should build? One that is repetitive, high volume and already measured, such as enquiry triage, first-pass quotes, invoice coding or call summaries into the CRM. The best choice is the one where you can state today's cost in hours or turnaround time.
How long does it take to embed AI in one process? A typical first process takes a few weeks to set up, then 30 days of measurement against the starting number. Most of the time goes into mapping the steps and connecting the system of record.
Sources
- Aussie businesses risk falling behind in AI race, CPA Australia media release, 9 October 2026, for the headline figures and the Gavan Ord quote.
- Business Technology Report 2026, CPA Australia, for survey method, integration levels by market and size, and the governance figures.
- Business Technology Report 2026: Australia infographic, CPA Australia, for the Australian figures on agentic AI, custom tools and oversight.
- AI adoption insights: December 2025 to February 2026, National AI Centre, for SME adoption and the rise in broad adoption.
Want help picking the first process and getting it running on AI properly? See our AI consulting work, or talk to us.
Is your business in the 88% or the 10%?
In 15 minutes we will look at one process and tell you what it would take to run it on AI.