Tech Council says only 5% of Australian SMEs use AI broadly

In its October 22, 2026 research release, the Tech Council of Australia said roughly one in four Australian small and medium-sized enterprises uses AI and only 5% uses it broadly across operations; chief executive Dr Kate Cornick called the distance between current use and potential value “a significant gap”. Delwood Advisory undertook the study with support from Employment Hero, Google and Deputy, using data contributed by Deputy, Xero and LinkedIn.

Google’s account of the launch describes a modelled opportunity of up to A$93 billion in additional annual economic value if adoption expands over the next five years and A$136,000 in annual productivity value for an average SME moving to broad use; it also said Google would deliver workshops at a Tech Council event in Sydney on October 23. The money figures are possible outcomes of wider adoption, while the usage figures describe current uptake.

Adoption varies by place and industry

Deputy’s launch summary puts AI use at 16% of regional SMEs versus 30% of metropolitan SMEs, and 33% of services SMEs versus 16% in construction; it labels a separate finding that 94% of AI-using SMEs reported productivity improvements as international research, and lists administration, rostering, marketing, customer communication, websites, quoting and invoicing among the Sydney program’s tasks. That overseas result is self-reported experience among users, rather than a measured gain for the Australian businesses counted in the adoption figures.

The location and industry gaps point to different conditions for putting a tool into regular work. A regional operator may have less access to specialist advice; a construction business may spend time on quotes, while a hospitality business may need to adjust rosters around demand and absences. Limited understanding, skills gaps, uncertain returns and concerns about security and reliability also appear among the barriers identified in the release. Those obstacles affect whether an initial experiment becomes a dependable process.

What the productivity forecast represents

The average-firm estimate models a move from no AI use to broad adoption. It represents potential annual productivity value under that scenario, rather than cash already saved or a return available from buying a single tool. The national figure is an estimate of additional annual economic value under wider uptake, over the report’s forecast adoption period; it is not realised revenue.

That distinction matters to a firm that already uses AI for an occasional draft. Producing marketing copy faster may save part of one employee’s day, but the larger operational benefit depends on what happens afterward: whether an enquiry is answered, a quote is approved, work is scheduled and an invoice is issued with less delay or correction. A quick first step can still leave the same bottleneck in the next handoff. The modelling concerns a change across recurring operations, not the value of an isolated prompt.

From one task to connected work

The named business activities make it possible to describe a progression from occasional use to wider adoption. The stages below are an editorial map of the work, rather than separate survey categories.

1. Single task. AI helps draft a customer reply, refine website wording or prepare the first version of a quote. A person checks the output and completes the surrounding work. The immediate question is whether that bounded task takes less time at an acceptable quality.
2. Connected workflow. One completed step feeds another. An enquiry can become a reviewed quote, a booking or roster decision, and then an invoice. The time spent checking details, correcting mistakes and moving work between people belongs in the assessment alongside drafting time. A quicker quote that creates extra billing corrections has shifted work rather than removed it.
3. Broad operational use. Several recurring workflows use assistance where it helps, with staff still responsible for approvals and exceptions. Scheduling, customer communication, administration and billing then change as part of normal operations. This is the relevant scale for understanding the small broad-use share in the research and its modelled productivity scenario.

The progression does not require every function to use AI. A connection is useful when information produced in one part of the business can be carried into the next without duplicating entry or losing a necessary human decision. It also exposes where an apparently successful experiment stops at the boundary of another system or team.

A measurable starting workflow

For an SME deciding where the reported opportunity might apply, a recurring workflow with a clear beginning and completed outcome is more informative than a collection of unrelated prompts. Quoting, rostering and invoicing are plausible candidates because each has a result and can involve handoffs. The choice still depends on where the firm actually spends time and where errors or delays recur.

• Set the start and finish, such as receiving an enquiry and approving a quote.
• Record current staff time, turnaround time and corrections for completed work.
• Identify a bounded step for AI assistance and the person responsible for checking its output.
• Compare the same measures after a change, including rework and any added tool cost.

This is a small operational test, not a way to claim the modelled average for an individual business. If faster drafting is cancelled out by extra checking, the workflow has produced no demonstrated saving. If the gain survives the handoff and repeat work, the business has evidence for extending that approach to another process.

Also read:

• Claude Adds Official Voting Links—Political Chats Are 5.75% of US Use
• LangGraph vs CrewAI: Control Pays Off Only When Agent Workflows Get Complex
• Americans use ChatGPT at work most often to make decisions