Why most Australian businesses aren't getting value from AI yet

As at November 2026, few Australian businesses can show a financial return from AI. In PwC's 2026 Global CEO Survey, 14% of 108 Australian CEOs said their company had seen revenue gains from AI, against 30% globally, Information Age reported. In 2024-25, only 15.8% of businesses with 20 to 199 staff measured what their digital activities contributed, the ABS found. The evidence points to the groundwork more than the tools: work nobody has measured, staff untrained on real tasks, and pilots that never become routine.

Are Australian businesses getting value from AI?

Mostly not yet, if value means money. PwC surveyed 4,454 CEOs in 95 countries and territories between 30 September and 10 November 2025. Information Age reported on 23 January 2026 that 14% of the 108 Australian CEOs said their company had seen revenue gains from AI, against 30% globally. In the same report, only 28% of Australian CEOs said they had invested enough to meet their AI goals, against 40% globally.

With 108 CEOs, each answer moves the Australian result by almost one percentage point, so the gap is indicative. PwC weights its global figures by GDP but not its country figures.

Globally, most CEOs have not seen a return either. In PwC's 2026 Global CEO Survey report, 30% of CEOs said AI had lifted revenue. More than half (56%) had seen neither higher revenue nor lower costs, and only 12% had seen both. On costs, 26% said AI had cut them and 22% said it had pushed them up, Search Engine Journal reported. Most of the CEOs surveyed run large organisations: only 30% lead one with revenue of up to US$100 million.

Smaller businesses report some benefit, but little of it reaches the bottom line. In NAB's SME Business Insights report of July 2026, SMEs using AI scored the improvement to their operations at 4.4 out of 10 on average, in the first quarter of 2026. Only 4% saw no benefit at all. Staff productivity (58%), marketing (46%) and customer service (36%) were most often named as the areas helped most. Only 9% named profitability, and 7% revenue. NAB says the benefits are often indirect, such as time saved, and harder to put a dollar figure on in smaller firms without formal measurement.

Time saved is not yet money made. McKinsey's State of AI in 2026 survey had 1,719 respondents worldwide, Fortune reported. According to FM Magazine, 80% said AI had improved their individual productivity. Only 37% attributed at least some impact on earnings before interest and tax (EBIT) to AI, about the same as a year earlier. An hour saved shows up in the accounts only if it changes what the business spends or earns.

How many Australian businesses use AI at all?

It depends on the measure. The ABS Business Characteristics Survey, released on 25 June 2026, covered nearly 7,000 businesses. It found 22.3% of businesses with 20 to 199 staff used AI in 2024-25, and 12.3% of those with 5 to 19. The ABS media release puts medium-sized businesses at 3% in 2021-22.

The ABS counts AI as one item in a list of technologies used during the year. The figure shows whether a business used AI. It says nothing about how well AI was used or what it returned.

The National AI Centre's AI adoption tracker measures something else. An average of 43% of SMEs reported some level of AI adoption between December 2025 and February 2026, according to Front Row Tech's August 2026 summary of the tracker. The tracker surveys about 400 SME owners and decision-makers a month. It counts any level of use the owner describes, including limited or one-off use.

The two answer different questions. The ABS asks whether a business used AI during a financial year. The tracker asks owners to describe their own adoption. Neither measures whether AI pays off.

Why isn't AI paying off yet?

The research points to four gaps. Experience from enterprise software adds a fifth.

Nobody measured the work first

Only 15.8% of businesses with 20 to 199 staff measured what their digital activities contributed to performance in 2024-25, according to the ABS. About a quarter (26.1%) actively collected or analysed data to make decisions. And 21.4% said uncertainty about costs and benefits had limited their use of technology.

Without a baseline, nobody can say what changed. If no one timed supplier invoice entry before Copilot arrived, any saving is an estimate. Treasury's evaluation of its own Copilot trial noted that its self-reported data may carry reporting bias.

The pilot has no owner

In Deloitte's State of AI in the Enterprise survey, released in February 2026, 28% of Australian respondents had moved 40% or more of their AI pilots into production, against 25% globally. Most had not.

Treasury's 2024 Copilot trial shows what can happen inside a pilot. Of 218 staff given Copilot for 14 weeks, most used it 2 to 3 times a week or less. Some gave up when it fell short of their expectations. Its clearest benefits were in basic administrative tasks, such as summarising information and drafting. The Australian Centre for Evaluation found no evidence it had improved workflows in the short term. Its first recommendation was to hand out licences against clear, specific use cases.

When everyone gets a licence and is told to try it, nobody is accountable for a process getting better. One named person who owns one process, and its numbers, can tell you each month whether it has.

The data and systems aren't ready

In Deloitte's survey, 42% of Australian companies pointed to cost hurdles and to technology or data availability. Treasury found Copilot performed worse than AI products staff had used elsewhere, partly because of restrictions in its IT security environment.

An AI tool can only draft from, match against or sort what is in your systems. If supplier details sit in three spreadsheets and an inbox, the tool can work from the wrong one. For accounting software, see what AI can do with Xero and MYOB right now.

Staff weren't trained on their own tasks

In the University of Melbourne and KPMG's 2025 study, only 24% of Australians had formal or informal training in AI or related fields. Among Australian workers, 57% had relied on AI output at work without checking its accuracy. And 59% had made mistakes in their work because of AI.

NAB found only 17% of SMEs rated their workforce as highly prepared for AI. Among businesses with 20 to 199 staff, 22.6% told the ABS that insufficient staff skills limited their use of technology. Treasury's trial participants often asked for more tailored training.

In a Danish study of about 25,000 workers, the reported benefits of AI chatbots were 10% to 40% greater where employers encouraged their use. The training also has to fit the work: a general course won't show your accounts team how to check a supplier match.

Nobody keeps the tools current

The evidence here is thin. Treasury's evaluation comes closest: it recommends monitoring AI's effects over the longer term, and periodic checks on whether newer products would suit better. No study found for this article measures what happens when nobody does this.

The rest comes from enterprise software. At DXC Technology, Coverleo's founder led AI-driven testing of business processes for about 40 Oracle Cloud clients at each quarterly update. Each update arrived on Oracle's timetable, and each time someone had to check that the client's processes still worked. AI tools change on their vendors' timetables too. Someone in your business has to notice, and retest the processes that depend on them.

What do businesses getting returns do differently?

They have done more of the groundwork. In PwC's survey, the one in eight CEOs reporting both higher revenue and lower costs were furthest ahead on what PwC calls AI foundations. PwC lists four: a technology environment that lets AI work across the business, a clearly defined road map, formal responsible AI and risk processes, and a culture that supports adoption. Its global release says CEOs whose organisations have strong AI foundations are three times more likely to report meaningful financial returns. PwC's report also says "isolated, tactical AI projects often don't deliver measurable value".

In Fortune's report on the McKinsey survey, high performers are about 6% of organisations and attribute at least 5% of EBIT to AI. Nearly three-quarters of them have fundamentally redesigned workflows, up from 55% a year earlier.

BCG's 2025 study of more than 1,250 firms found 5% achieving AI value at scale, and 60% reporting minimal revenue and cost gains. BCG's 10-20-70 rule puts 70% of the focus on people and processes, 20% on technology and 10% on algorithms.

All three are self-reported surveys, mostly of large companies. They show which practices go with returns. They cannot show that the practices cause them.

Is the "40% faster" figure real?

It is real for the tasks tested, and those tasks were short. In a 2023 experiment published in Science, Shakked Noy and Whitney Zhang gave 453 college-educated professionals writing tasks from their own occupations, such as press releases, short reports and delicate emails. Each task took 20 to 30 minutes. Half the group was randomly given ChatGPT. Their time fell by 40%, and the quality of their work rose by 18%, graded blind by professionals.

The authors named the limit themselves. The tasks were short and self-contained, and needed little context-specific knowledge. In their working paper, they wrote that this "may inflate our estimates of ChatGPT's usefulness". The tool was 2023's ChatGPT.

Real jobs look different. Anders Humlum and Emilie Vestergaard linked surveys of about 25,000 Danish workers in 11 occupations, including accountants, office clerks and customer support staff, to employment records. Users reported average time savings of about 3% of their work hours. The authors found no significant effect on earnings or hours worked.

A 30-minute writing task with clear instructions is a different job from a month-end close. No Australian study found for this article measures AI time savings in small-business back offices. So neither number is your saving. Measure your own.

What should a business of 10 to 200 staff fix first?

Work through these in order.

1. Time three recurring processes before you change anything, such as supplier invoice entry, bank reconciliation or customer enquiries. Record volumes, time taken, errors and rework.
2. Name one owner for each process. That person reports on it every month.
3. Approve one tool, and write down what data may go into it.
4. Train each team on its own tasks, with real invoices and emails, and show them how to check the output. AI tools prepare: they draft, match, sort and flag. People send, approve and pay. If software helps make decisions about people, read about the Privacy Act rule on automated decisions that starts 10 December 2026.
5. Measure each process the same way every month. Count volumes, errors and rework as well as time, because self-reported savings can be biased.
6. Review the tools each quarter, and retest the processes that rely on anything that changed.

Does ongoing support work better than a one-off project?

The evidence supports sustained effort. It does not say who should provide it.

Each of the stronger findings above describes work that carries on after launch. Danish workers reported bigger benefits where employers encouraged use. McKinsey's high performers redesigned their workflows. Treasury recommended "sustained investment and effort", with monitoring over the longer term.

No study found for this article compares ongoing support with a one-off consultant or a new hire. For the three options side by side, see AI consultant, in-house hire or ongoing support.