An aggregate figure for 'what Australian small businesses spend on AI' averages together a sole trader paying $30 a month for a Claude subscription and a 150-person firm running a six-figure annual AI budget across multiple workflows. That average is true and useless at the same time, because neither business can learn anything from it. What actually helps is a benchmark broken out by size band, so a business owner can compare their spend against something genuinely like their own operation.
Why size band matters more than industry
AI spend correlates more tightly with headcount and workflow maturity than with industry sector. A 20-person accounting firm and a 20-person marketing agency, despite doing very different work, tend to land in a similar AI cost range once you account for seat count and the number of workflows each has actually built out, because the cost drivers, subscription seats, API usage volume, and setup investment, scale with team size more than with what the business actually sells.
The 2026 size bands
Based on the pattern we see across Australian SMB clients and comparable published data, four size bands cover most of the market with meaningfully different cost profiles.
Sole trader to 3 staff: typically $50 to $400 a month, usually one or two subscriptions plus light API usage, minimal dedicated build spend.
5 to 20 staff: typically $600 to $3,500 a month, a mix of team subscriptions and one or two owned workflows, often $2,000 to $8,000 in one-off setup spend for the first Cowork skills.
20 to 50 staff: typically $3,000 to $9,000 a month, several owned workflows, a named AI budget line, and setup investment in the $8,000 to $20,000 range across the year.
50 to 200 staff: typically $8,000 to $30,000 a month, dedicated internal ownership of the AI stack, and annual build and maintenance spend often exceeding $30,000.
Where businesses commonly sit outside their band, and why
A business spending well above its band's typical range is usually either running redundant subscriptions nobody has audited, or has genuinely invested heavily in owned workflows that are paying back at a rate that justifies the spend, worth distinguishing before assuming either is a problem. A business spending well below its band, meanwhile, is often leaving real time savings on the table, still doing manually what a modest investment would automate, rather than being admirably frugal. The benchmark is a starting point for the conversation, not a target to hit exactly.
A worked example: a 34-person Brisbane engineering firm
A 34-person Brisbane engineering consultancy reviewing its AI spend against these bands found itself at roughly $11,200 a month, above the top of its expected range for its size. Auditing the spend found the gap was explained by three overlapping report-formatting subscriptions covering essentially the same workflow, a legacy of different project teams adopting different tools independently over eighteen months without central coordination. Consolidating onto one Claude Cowork-based workflow brought monthly spend down to roughly $6,800, back inside the expected band for a firm its size, without losing any of the underlying capability.
Using the benchmark without over-relying on it
These figures are a sense-check, not a budget target. A business with unusually high-volume, high-value AI-assisted workflows can reasonably sit above its band and be getting excellent return on that spend. The value of the benchmark is catching the businesses that are outside their expected range for reasons nobody has actually examined, redundant tools, an unmetered runaway workflow, or simply not knowing what peers of a similar size typically spend.
Where the bands are likely to move next
Two forces are pulling on these numbers as 2026 progresses. Underlying model pricing has generally continued to fall, which should push the mid and upper bands down over time for businesses running mostly API-based owned workflows. At the same time, more Australian SMBs are moving from a single subscription to genuinely owned workflows, which pushes one-off setup spend up even as ongoing run costs trend down. The net effect for most businesses in the 20 to 200 staff range is likely a shift toward more upfront build spend and lower ongoing per-task cost, a healthier mix than a flat subscription bill that only ever climbs.
Reviewing your own numbers against the band closest to your headcount once a year, alongside the usual budget review, is a quick way to catch drift before it becomes a much larger number. For a business that has never run this comparison, the first pass is usually the most revealing, because it is the first time anyone has looked at total AI spend as a single number rather than a scattered set of individual subscription line items.
