The golden rule
Runs create proposals for you to approve — nothing sends, changes, or spends until you (a) approve it, or (b) raise the Autonomy dial. You're always in control.
1 Schedules & triggers — when it runs
Add a trigger to decide when the AI runs:
- On a schedule — e.g. a morning/nightly run.
- When a new lead arrives — it reacts the moment an enquiry lands.
Example: add a "when a new lead arrives" trigger so every enquiry is automatically qualified and a draft reply is prepared — waiting in your approval queue when you open the admin.
2 Autonomy dial — how far it can act
This is the trust setting. Start low, raise it as you gain confidence:
- Supervised — every run just proposes; you approve everything. (Recommended to start.)
- Hands-off — clean, unflagged runs act on their own; anything unusual still waits for you.
3 Live actions vs dry-run — whether it really sends
Live actions · off = dry-run ($0), on = real actions
Left off, automations show you exactly what they would do without doing it — free, no risk. Turn it on when you're ready for real sends/changes. Real actions (like sending) need an action webhook set, and — for live AI — may use metered API credit.
If "live" is on but no notify webhook is set (Automations → action webhook), quotes/messages generate but won't actually send. Set the webhook to complete the loop.
4 The kill switch & the ROI report
- Kill switch — one control to stop all automated actions immediately if you ever need to.
- "What your autopilot did" — a running report of every automated action and the time/$ it saved, so you can see the value and prove it to a client.
A complete example
"Job done → review request." When you mark a lead Job done in the Pipeline, an automation drafts a friendly review request for that client. It appears as a proposal; you glance at it and approve — or, with the dial raised and live actions on, it sends on its own. Set it up in Automations by enabling the "on job done" trigger.