Small business
Automation that makes decisions, on data you keep.
Start with one workflow and grow to a full platform without rebuilding.
What you get
This path is for a smaller team that wants one workflow automated properly, not a slide about a future platform. The same practice that designs AI factories runs this work. The first system is small enough to finish, and it is built so a later platform does not throw it away.
Routine work off your team
A workflow you already run, taken off the people who repeat it, so their time goes to the work that needs them.
Bounded decisions
Decisions stay inside a limit you set, with human sign off, approvals and audit logs. A person can stop the agent.
Data kept in Australia
Your data stays in Australia, under your control, on a hosted service, a private cloud, or on premises.
The neighbouring practices are Automation for the repeatable path, and Agentic AI when a step needs judgement rather than a fixed rule. Integration with the systems you already run is AI integration.
How it runs
Four steps. You can stop after any one of them. The next step does not rebuild the last.
01
Audit
Name one workflow, the data it uses, and the decisions a person must still make.
02
Pilot
Run that workflow on real work, inside the boundary the audit set.
03
Run
Keep it in daily use, with approvals, an audit log, and a person who can stop it.
04
Scale
Add another workflow, or the platform around the first, without rebuilding it.
Packages
Four ways to engage. Durations are shown. Prices are not listed, because cost is scoped after the audit.
Automation Audit
2 weeks
A look at one workflow: what can leave the team, what must stop for a person, and the data the work depends on.
Pilot
6 to 8 weeks
Run that workflow inside the boundary the audit set, with an owner, approvals, and a record of what ran.
Managed Operations
Monthly
An operating rhythm after the workflow is in use: monitoring, changes, and a named way to handle a decision that must stop.
Scale
By consultation
When one workflow should become a platform, or when a later workflow should join the first without a rebuild. Scope is agreed for that step.
Models and where they run
The model follows the workflow. We do not start by training one. Deployment is a choice about control, not a reason to redesign the decision.
- Retrieval
- The usual start. The model answers from sources you have approved, and the record shows which source was used. This pattern is often called RAG.
- Fine tuning
- Used when the audit shows a general model misses your language or your cases. It is not the default.
- Pre training
- Only where it pays. A model trained from scratch is a different project, and it is not how one workflow starts.
- Deployment and support
- Hosted, private cloud, or on premises. In each case the data stays in Australia and under your control.
The public figures
In the 2024 to 2025 financial year, the Australian Bureau of Statistics reported that uptake of artificial intelligence among small and micro businesses was around 11 per cent. In that survey a small business has 5 to 19 employees and a micro business has 0 to 4. The figure is the two groups together. It is reported use, not a benchmark we set.
Australian Bureau of Statistics, business adoption of artificial intelligence, 2024 to 2025. Size groups are defined in the Business Characteristics Survey methodology for 2024 to 2025.
A note on privacy transparency
From 10 December 2026, the Office of the Australian Information Commissioner says an organisation’s privacy policy needs to cover certain automated decision making. That is where the organisation has arranged for a computer program to make, or to do something substantially and directly related to making, a decision that could significantly affect a person’s rights or interests, and personal information is used.
This note points at the Commissioner’s own explanation. It is not legal advice, and it is not a statement of what your policy must say.
OAIC: new resources on transparency for use of AI and automated decision making
Briefings in series
Answers
Direct answers
Short statements for the questions this page is here to settle.
What does small business automation cost?
Cost is scoped after the audit. This site does not publish a price. The audit names the workflow, the boundary and the work a pilot would take. Any figure is agreed in writing before that work starts.
What do you need from us to start?
We need one workflow you can describe, the person who owns the outcome, and the systems and data that workflow already uses. We also need to know which decisions may proceed and which must stop for a person. You do not need your own GPU cluster to start.
What happens when the agent is wrong?
The agent has an owner, a limit on the actions it can take, and a record of the steps it took. A decision that changes money, access, or a record about a person stops for a person. When an outcome is wrong, the owner can reject it, and the path that does not use the agent stays available.
Who owns the data and the models?
You keep your data. It is held in Australia under your control, on a hosted service, a private cloud, or on premises. We do not train a model on your records unless that is agreed in writing. Where a model is fine tuned or built for you, who owns that model and the configuration is set in the agreement. A foundation model from another provider stays under that provider's terms.
Book a 30 minute automation audit call
Bring one workflow, the person who owns it, and the systems it already touches. We will tell you whether an audit is the right next step.