The headline ranges
| Stage | What it is | Typical cost (AUD) |
|---|---|---|
| Agent mapping | Process, risks, scorecard, fixed pilot quote | A$4,500 (credited) |
| Pilot | Working agent on your data, 4 weeks, eval harness | A$12k – A$30k |
| Production agent | Hardened, integrated, governed, dashboard | A$30k – A$120k |
| Multi-agent system | Orchestration, shared memory, controls | A$120k – A$250k |
| Run costs | Models, platform, monitoring, improvements | A$300 – A$3,000/mo + managed ops from A$2k/mo |
What moves the price
- Integrations. Each system the agent reads or writes adds work — and the old ones without APIs add the most.
- Decision scope. The wider the set of actions it may take, the more guardrails, tests and approvals are needed.
- Evaluation. A credible eval set takes real cases and real time to build; it is also what makes the agent trustworthy.
- Governance. Privacy review, access control, logging and escalation design — cheap to include early, expensive to retrofit.
- Volume. Run cost scales with tasks; we set a monthly ceiling so there are no surprises.
The hidden costs
- Data clean-up before the agent can work (often the first A$5k–A$15k of any project).
- Change management — the people whose work changes need a plan, not a memo.
- Model price changes — use a model-agnostic architecture so you can switch.
- The pilot that never ends — insist on a scorecard and a decision date.
How to compare quotes
Ask for the eval metrics the vendor will commit to, the governance file, the run-cost ceiling and who owns the code. A quote without those four is a demo, whatever the number says.
Questions people ask
Is a pilot worth it if we already know what we want?
Yes — it is the cheapest way to find out whether the agent can hit the number on your data. A failed A$12k pilot beats a stalled A$100k build.
Do run costs really stay under A$3k a month?
For most single agents, yes. High-volume support or voice agents can exceed it; the pilot measures cost per task so you know before production.
Related: AI agents · Agent vs automation vs chatbot · AI Readiness Audit