Approval guide
Human-in-the-Loop AI Workflow Guide
Use this when AI prepares work but staff must control system-of-record changes.
What should require review
Read-only summaries can run automatically. Review system writes, messages, customer updates, and published content.
- Planning Center, ministry task, volunteer, event, or church website updates.
- Nonprofit program, donor, volunteer, grant, or outreach follow-up.
- CRM updates, lead changes, and sales follow-up.
- Invoices, estimates, payment notes, or finance follow-up.
- Calendar changes and task creation.
- Website, campaign, email, or communication drafts.
How to design an approval path
Show the proposed change, reason, sources, and reviewer without slowing every low-risk action.
- Show the proposed change before it runs.
- Make the source context visible enough to verify.
- Separate approve, reject, and edit decisions.
- Keep action history for later review.
What happens after approval
Define what follows approval, denial, or failure. Nurii runs an approved action and continues the thread with the next dependent step.
- Approval can unlock a second action that depends on the first action's result.
- Denial can send the agent back to revise, ask for clarification, or stop.
- Failure can trigger a retry, a wait step, or a safer alternative proposal.
- Stop controls and undo support should be visible where active or executed work can be interrupted.
Match review to risk
Match review to risk; a read-only report needs less control than a finance update or public email.
- Low risk: summaries, dashboards, team-only drafts, and reminders.
- Medium risk: tasks, status updates, team follow-up, and CRM notes.
- High risk: payments, public content, customer messages, and system-of-record changes.
- Always define who owns each approval decision.
Approval makes AI easier to trust
Review makes operational AI useful by preventing important changes from running silently.
- People can inspect proposed work before execution.
- Admins can keep sensitive systems protected.
- Teams can start with review and automate more as trust grows.
- Action history gives operations teams a trail to inspect later.
Checklist
Approval design checklist
Which actions can run automatically after advance approval?
Which actions always require review?
Who can approve each type of action?
What source context should reviewers see?
What should happen after approval, denial, or failure?
Where should action history be visible?
Evaluation guide
AI Operations Platform Guide
Learn how to evaluate AI operations platforms for connected apps, dashboards, AI chat, workflow automation, approvals, and team permissions.
Read guideWorkflow guide
AI Workflow Automation Guide
A practical guide to AI workflow automation for scheduled reports, event-triggered workflows, monitoring, Slack and email delivery, and reviewed follow-up.
Read guideChecklist
AI Integration Planning Checklist
Plan which apps, accounts, files, calendars, dashboards, access rules, and approval steps should come first in an AI operations setup.
Read guideReady to apply this guide?
Map the first workflow, sources, and approvals.