AI adoption in the Australian mid-market: what the numbers actually say

Australia's own statistical agency says 22% of medium-sized Australian businesses used AI in 2024-25. Vendor surveys of roughly the same market report 60 to 68%. Both cannot be right, and the gap between them is the most useful number in Australian AI right now, because it is the difference between using a chatbot and running a system.

We went looking for research on Australian businesses of 50 to 500 people, which is the segment where most of the country's revenue teams sit. What we found was a pile of numbers that flatly contradict each other. Here is what the credible sources actually say, what the contradiction means, and which figure to quote in a board paper.

Method, up front

Stating this first because it changes how much weight to give what follows.

• What this is: a synthesis of published research, not our own fieldwork. We did not run a survey. Every number below is sourced and linked so you can check it.
• Primary source: the Australian Bureau of Statistics, on nearly 7,000 businesses, fieldwork October 2025 to February 2026, measuring the 2024-25 financial year.
• Secondary: MIT's The GenAI Divide: State of AI in Business 2025, plus CSIRO and Department of Industry figures.
• The bias to know about: we sell AI and RevOps engineering to exactly the businesses in this data. One finding below happens to flatter firms like ours, and we have flagged it where it appears rather than burying the disclosure.

Finding one: adoption is far lower than you have been told

The ABS number is the one to quote, because it is the largest sample, the clearest method, and nobody is selling anything with it.

Medium-sized businesses went from 3% to 22% in three years. That is a sevenfold increase and it is genuinely fast. It is also nothing like the 60 to 68% that circulates in vendor decks, and one of those numbers is going into your board paper.

Filter the ABS data to innovation-active medium businesses and it rises to 28%. Still not 68%.

Finding two: the gap is a definition, not a lie

The contradiction resolves once you notice the surveys are measuring different things.

Ask "does anyone here use AI" and you capture one person using ChatGPT to rewrite an email. Ask "has this business adopted AI" the way a statistical agency does and you capture something closer to a system. Almost every survey blurs the two, which is why adoption numbers are always implausibly high, and it is why your competitors sound further ahead than they are.

The useful question is not whether your business uses AI. It is whether anything runs without a person triggering it. On that definition almost nobody in the Australian mid-market has adopted AI, including plenty of businesses that will tell you they have.

Finding three: most of it produces nothing

MIT's 2025 study reviewed 300 public deployments alongside 52 executive interviews and 153 leader surveys. The headline: 95% of generative AI pilots delivered no measurable profit and loss impact. Over 80% of organisations had piloted something like ChatGPT or Copilot, and around 40% reported a deployment, but the value mostly landed as individual productivity rather than a business outcome anybody could point to.

Read alongside the ABS data, the picture for a 150 person Australian business is: roughly a one in five chance you have adopted AI at all, and if you have, a one in twenty chance it is doing anything your CFO can see.

Finding four: who actually succeeds

The same study found 67% of externally partnered deployments succeeded, against 33% of internal builds.

We are an external partner, so treat that number with the scepticism it deserves coming from us. The mechanism is worth more than the statistic anyway, and it is not that agencies are clever. It is that an external engagement has a scope, a deadline and an owner, and an internal project competes with everyone's day job until it quietly stops. You can get the same effect internally by naming an owner and giving them time. Most businesses do neither.

Finding five: the money is in the wrong place

Between 50 and 70% of AI budgets go to sales and marketing, while back-office automation delivers the clearer return. That is the most actionable finding in the whole set, and it is the one nobody acts on, because sales and marketing AI demos well and invoice processing does not.

What we think it means

Three things, clearly labelled as our reading rather than the data.

You are less behind than you think. If you are a 150 person Australian business with no AI in production, you are with the majority, not the stragglers. Decisions made from a sense of being behind are usually bad ones.

The bottleneck is ownership, not technology. A 95% failure rate against tools this capable is not a capability problem. It is what happens when nobody owns the thing after the pilot impresses everyone in the demo.

The winning move is boring. One back-office process, one named owner, one number you measure before and after. That is the whole method, and on this data it puts you ahead of roughly 95% of the market.

What nobody has measured

Being straight about the limits of this, since that is the part usually left out.

The ABS measures a financial year that ended over a year ago, and in this category a year is a long time. Nothing here segments Australian mid-market businesses by function, so there is no reliable read on what revenue teams specifically are running. The MIT sample is enterprise-weighted and not Australian. And the failure figure counts measurable profit and loss impact, which understates genuine but unmeasured gains.

Which is exactly the hole we intend to fill. We are running our own survey of Australian businesses between 50 and 500 people, and this article is the baseline it will be measured against. If you want the results when they land, or want to be part of the sample, tell us.

Use this

Every figure here is linked to its source. Quote them, put them in your board paper, and cite the original rather than us. If you are writing about Australian AI adoption and want the working, get in touch and we will send it over.