The Tech Council of Australia published a report on AI use by small and medium enterprises on October 22. Delwood Advisory conducted the analysis with support from Employment Hero, Google and Deputy. This is industry-supported research about Australia, and its findings need to be read in that context.
The authors model up to $93 billion in additional annual economic value from broader adoption over the next five years. This estimates potential rather than profits already realised. The figure is not a forecast of an individual company's revenue and cannot be transferred directly to Bulgaria.
A new publication using older inputs
The full report uses 2025 data for adoption indicators. Reported productivity and satisfaction benefits come from different cited studies. The new report's date therefore does not mean every number was measured in October 2026 across the same group of businesses.
That distinction matters when assessing the evidence. Early adopters' self-reports, a modelled economic scenario and measurement of a particular process answer different questions. Combining them in one document does not create a single experiment proving the same effect everywhere.
Google also describes practical training around the research launch. Technology providers' involvement is disclosed; we do not describe it as independent confirmation of their own product promises. The useful business question is how to check whether the ideas apply before relying on the forecast effect.
What a small-business test could look like
Our editorial proposal is to begin with one repeatable process that has an observable result and a human reviewer. Drafting a response to a standard enquiry is one example. Record the initial completion time, common errors and cases requiring further conversation with the customer.
After introducing an assistant, measure the entire route to an accepted response, including editing. Quickly producing a draft is insufficient if verification and corrections take longer. The test should compare completed work with completed work on tasks of similar complexity.
A separate record should cover data access, subscriptions, configuration and employee training. This lets the owner see what the chosen process actually requires. The method is SIExpo's proposal, rather than a company case in which we have measured a positive outcome.
When expansion is justified
Before further automation, establish who accepts the output and how the previous working method can be restored. Tests should include unsuitable, incomplete and contradictory enquiries. Otherwise, the measurement will cover only easy cases presented in a convenient demonstration.
The conclusion is not that every company already has a guaranteed return. The report draws attention to broader adoption opportunities, but a specific decision requires local data and a business's own assessment. For Bulgarian readers, the most transferable issue is how to organise that examination without presenting Australian economic scale as a promised local result.
*Cover: a conceptual AI illustration.*
*Sources checked: 2026-10-23T16:10:40.049366+00:00.*
