Days 1–5: choose the workflow
- List repeated work that creates delay, labour cost, errors or missed follow-up.
- Choose work with enough volume to measure and enough stability to repeat.
- Record the current time, quality, error rate and downstream business result.
- Write the smallest useful outcome for a 30-day pilot.
Days 6–10: design the guardrails
Define permitted inputs, prohibited data, approved tools, the human reviewer and when the output must be rejected or escalated. For customer-facing or consequential work, the control should be stronger than for an internal draft.
Use the AI policy starter to establish shared rules. For personal information, consult the OAIC’s current guidance.
Days 11–24: run a controlled pilot
| Record | Why |
|---|---|
| Manual minutes | Creates the real baseline |
| AI-assisted minutes | Shows capacity returned |
| Rework and errors | Prevents false efficiency |
| Business outcome | Connects activity to value |
| Exceptions | Reveals where the workflow breaks |
Days 25–30: decide and document
Scale when the workflow reliably improves the chosen measure and the controls are workable. Change it when the failure is specific and fixable. Stop when benefits are weak, review cost removes the gain, or the risk is not justified.
Your finished pilot should leave five artefacts: baseline, approved workflow, example inputs and outputs, review checklist, and a short result note.