What does AI actually cost a small business in Australia?
Software is the small line. Expect roughly $20–$40 per person per month for assistant licences, usage-based API costs that scale with volume, one-off integration work if AI must touch your existing systems, and internal time for whoever defines and checks the work. The last line is the one most budgets omit and most overruns come from.
The four cost lines
Every AI project an SME runs has the same four cost lines. They behave very differently from one another, and confusing them is why quotes vary so wildly.
| Cost line | Typical shape | Scales with |
|---|---|---|
| Per-seat licences | Roughly $20–$40 per person per month for mainstream business AI assistants | Headcount |
| API / usage | Fractions of a cent per request, but unbounded | Volume of work processed |
| Integration build | One-off project cost, quoted per scope | Number and age of systems involved |
| Internal time | Rarely costed at all | How poorly the task was defined |
Per-seat licences are predictable and easy to cancel, which makes them the right place to start. Usage-based costs are the opposite: cheap per unit, and capable of surprising you when a process that ran ten times a day starts running ten thousand times. Anything built on usage pricing needs a spend cap configured on day one, not after the first unexpected invoice.
Integration is where the money actually goes
If AI only has to read a document a person pastes in, integration costs nothing. The cost appears the moment it must read from or write to something you already run — a practice management system, an accounting package, a CRM, a shared drive with fifteen years of inconsistent folder names.
The price of that work is driven almost entirely by the systems, not the AI. A platform with a documented modern API is straightforward. A system with no API, or one that requires screen automation, or data spread across spreadsheets with no consistent identifier, can cost more to connect than everything else in the project combined. This is why a credible quote asks what you currently run before naming a number, and why a quote given without that question should be treated as a guess.
The line nobody budgets
Someone inside the business has to explain what good output looks like, review early results, and correct them. For a first project that is commonly several hours a week for a month or two, and it falls on your most experienced person — because they are the only one who knows the answer. Costed honestly at a senior internal rate, this is frequently the largest single line in a first AI project.
It is also the line that shrinks the most on a second project, which is why the first one always looks like poor value in isolation and why judging an AI programme on its first deployment is misleading in both directions.
What a sensible first commitment looks like
A reasonable first step is small enough that a disappointing result is affordable and informative. In practice that means licences for the handful of people doing the work, no integration build at all, a defined two-to-four week trial on one task, and a named figure you are trying to move. The deliverable is a decision, not a system.
If a proposal asks for an integration build before anyone has measured whether the task benefits, the sequence is backwards. The build is the expensive, hard-to-reverse part, and nothing yet justifies it.
How NTDG quotes it
NTDG quotes a fixed price per engagement rather than from a rate card, because the cost is driven by how many systems are involved and how well the process is currently documented — neither of which is visible from a day rate. Send an outline of the process and what you run it on, and you get a fixed figure and a scope back. Nothing here is financial advice; it is how the work is priced.
This question comes up most often in our AI Orchestration & Business Deep Dives engagements.