The honest range for a genuine Australian small-business AI automation, not a chatbot demo, but something that reliably does real work, sits between $3,500 for a narrow single-workflow build and $35,000 for a multi-system rollout across a whole back office. The number that matters more than the range is the shape of the spend: setup cost once, then a much smaller ongoing cost, versus the SaaS-subscription model most owners are used to.
What actually drives the cost
Three things move the number more than anything else: how many systems the automation needs to touch, how much judgement versus straightforward data-moving the task requires, and how much testing against real edge cases happens before it goes live. A workflow that reads one system and drafts a document is cheap. A workflow that reads three systems, makes a judgement call, and writes back into two of them is not, mostly because of the testing time, not the build time.
Single-workflow, one system: $3,500 to $6,000 setup, days not weeks
Multi-step, two to three systems: $8,000 to $18,000, one to three weeks
Full back-office rollout, five plus systems, multiple approval gates: $20,000 to $35,000, four to eight weeks
Ongoing running cost once built: typically $300 to $1,200 a month in subscriptions and model usage
Where the money actually goes
A useful way to think about it: roughly 30 percent of the cost is the initial build, 40 percent is testing against real messy data (the invoice that doesn't match the template, the customer who writes an email in three different formats), and 30 percent is the handover and documentation that lets your team run it without the builder on call. Vendors who quote a number without a testing-against-real-data phase are usually underquoting, and the gap shows up later as support tickets.
What a Brisbane trades business actually paid
A 15-person Brisbane electrical contractor automated job-costing summaries pulling from their scheduling software and Xero, a two-system build with a moderate judgement layer for flagging jobs running over budget. Total cost was $9,200, delivered over three weeks, with ongoing running costs of roughly $180 a month in Claude usage and connector fees. The owner estimated the manual version of this task, done by an office manager every Friday, cost around $260 a week in wages, meaning the build paid for itself in under nine months and kept saving after that.
The cheapest path that still works
The costs that catch owners off guard
Two costs get underestimated consistently. The first is the discovery phase, properly documenting how a workflow actually runs today, including the exceptions nobody mentions until week three, which can eat 15 to 20 percent of a project's total hours if the business hasn't already written the process down. The second is change management, the time it takes staff to trust a new automated step enough to stop double-checking it manually, which isn't a line item on any quote but genuinely affects how quickly the investment starts paying back.
A sensible way to budget is to add a 15 percent contingency on top of any fixed quote for a first automation project, specifically to cover discovery surprises, and to expect the first month of live running to involve more owner attention than months two onward, as trust in the system builds. Vendors who promise a completely hands-off result from day one are usually glossing over this settling-in period rather than eliminating it.
Finally, it's worth comparing this spend against the alternative most businesses default to without thinking about it: hiring. A part-time admin hire in most Australian cities runs $35,000 to $50,000 a year fully loaded for 20 hours a week of work. A $9,000 to $15,000 automation project handling a genuine slice of that workload, with a few hundred dollars a month in running costs, is a materially different financial commitment, and one that doesn't come with recruitment risk or the ramp-up time a new hire needs before they're fully productive.
A last practical note: get at least two quotes for any project over $8,000, and ask each vendor to walk through their testing phase specifically, not just the build. The gap between a vendor who tests against ten real historical examples and one who tests against a single happy-path demo is the gap between an automation that survives contact with your actual business and one that needs a rebuild three weeks after going live.
If $3,500 is still more than you want to spend before proving the concept, start with a single Claude Cowork skill built and tested by your own team over a weekend, zero build cost beyond the subscription, and use that to prove the workflow is worth paying to harden before commissioning a proper build. Plenty of Australian small businesses run entirely on self-built skills for six months before their first paid automation project, and that's a genuinely reasonable way to de-risk the spend.
