The cheapest genuine starting point for an Australian small business is a Claude subscription already covering the automation, plus one self-built Cowork skill, zero additional spend beyond what many businesses are already paying for AI access. That's a materially different starting point to the $3,500-plus figures that come up once a business commissions a paid build, and it's worth understanding the gap before assuming a paid project is the only entry point.
What genuinely costs nothing extra
A Claude subscription you're already paying for, most plans include Cowork mode and skill-building
One connector to a tool you already use, Gmail, Google Drive, or Notion, usually a five-minute authorisation
A self-written skill for one recurring task, built in an afternoon with no developer needed
Testing against real work for two weeks before trusting the output unsupervised
Where the free-to-cheap path actually breaks down
The zero-additional-cost path works well for tasks a business owner or a competent staff member can describe clearly and check the output of themselves. It breaks down once a workflow needs a custom connector to an internal system without an existing MCP server, needs to run unattended on a schedule with proper failure alerting, or needs enough testing rigour that getting it wrong has real financial consequences. That's the point where paying for a build starts making sense, not before.
A staged approach that minimises wasted spend
Stage one: pick the single most annoying repeated task in the business and build a skill for it yourself, unaided, over a weekend. Stage two: once that skill has run reliably against real work for a few weeks, decide whether it's worth hardening, scheduling, or connecting to another system, which is where a small paid project, typically $1,500 to $4,000 for a single hardened workflow, starts to make sense. Stage three, a genuinely multi-system automation, only becomes worth commissioning once stage two has proven the underlying pattern actually saves real time.
What a Wollongong tradie business actually did
A five-person Wollongong plumbing business started with zero additional spend: the owner used their existing Claude subscription to build a skill drafting job-completion summaries from voice-dictated notes, tested it himself for three weeks on real jobs before trusting it fully. Only after that proved reliable did he pay $2,200 for a developer to connect it directly to their job-management software rather than requiring a manual paste each time. Total cash spent before seeing genuine value: zero. Total spend once the value was proven: $2,200, a fraction of what a from-scratch commissioned build would have cost without that proof first.
The honest bottom line
What the free-to-cheap path teaches you before you spend real money
Beyond the direct cost saving, building the first skill yourself teaches you something a paid vendor conversation can't: exactly which parts of the task are genuinely mechanical versus which ones need real judgement, because you'll hit both while building it. That understanding makes the eventual paid-project conversation, if you get there, sharper and cheaper, because you can specify precisely what needs hardening rather than handing a vague brief to a developer and hoping they ask the right questions.
It's also the fastest way to find out a task isn't actually worth automating at all, some workflows look annoying in the abstract but turn out to be rare enough, or variable enough, that a manual approach remains genuinely more sensible. Discovering that during a free weekend experiment costs nothing. Discovering it after paying $4,000 for a build is a far more expensive lesson, and one worth avoiding by proving the pattern cheaply first.
One more genuinely free lever worth using before spending anything: Anthropic's own documentation and the growing library of publicly shared skills and prompt patterns mean a business rarely needs to design a workflow entirely from scratch. Searching for whether someone has already solved a similar problem, even approximately, and adapting it rather than starting blank often cuts the self-build time in half, which matters when the whole point of this path is minimising the time investment as well as the cash one.
The last thing worth saying plainly: cheap doesn't mean lower quality if the work is scoped honestly. A well-built self-serve skill covering a narrow, well-understood task can be just as reliable as a commissioned build covering the same narrow scope, the paid option earns its price on breadth, integration complexity, and testing rigour, not on some inherent gap in quality for the same-sized task.
If cost is the binding constraint, start with what you're already paying for and your own time, prove the pattern works on a task you understand well enough to check yourself, and only spend real money once that proof exists. Most Australian small businesses skip straight to asking 'how much does a build cost' when the genuinely cheapest and often more informative first step is building something small themselves first.
