Automate. Add automation and AI only where it earns its place, with a human at every point that matters.

What this phase is for

Automation is applied with purpose, to repetitive low-judgement work, and AI is added only where it creates real leverage. Human decision-making stays where it matters, with a named person at every review point. This phase comes fourth for a reason: automate a broken workflow and it fails faster.

Do it yourself

1. Automate the chasing first. Reminders, follow-ups, confirmations, "your turn" notifications. Boring, safe, and where most of the audited hours were.
2. Run the twelve-question assessment on every AI use. Before an AI drafts, sorts, scores, triages or answers anything that touches a client or a decision, run it through the assessment. Keep the answers with the tool register.
3. Name the checker. For each AI output that reaches a client or shapes a decision about a person, write the name of the person who reads it first. No name, no automation.
4. Keep decisions about people human. Pricing a customer, screening an applicant, declining a booking, escalating a complaint. Software may prepare these; a person decides. From 10 December 2026, if the Privacy Act covers you, your privacy policy has to say where software shapes decisions.
5. Write the bad-day page. Who is told, what is deleted, who decides whether a client is told. The first twenty four hours.
6. Start with one automation and watch it for thirty days. Nothing else ships until the first one has run clean for a month. Then add the next.

The AI lens on this phase

The test for every automation is one sentence: if this produced something plausible but wrong, who would notice before the client did? If the answer is nobody, it is not ready. That single question is most of what the government frameworks ask in sixteen.