Decision Tool
Human‑in‑the‑Loop Map
A map for defining where humans must remain—so automation supports judgment instead of replacing responsibility.
What this helps you decide
- Define the boundary between decision support and decision ownership.
- Prevent automation from silently taking responsibility away from a human.
- Design override points that match real operations.
When to use it
- You’re considering automating routing, approvals, communications, or “next actions.”
- Your work involves accountability, customer trust, or compliance constraints.
- You want automation that reduces mental load without removing human judgment.
The framework
Decisions AI should never finalize
- Judgment-heavy tradeoffs with real consequences.
- Relationship-based communication where tone and context matter.
- Moral or accountability decisions where a person must own the outcome.
Decisions AI can recommend on
- Summaries and suggested categories for a human to confirm.
- Draft responses that a person reviews and sends.
- Exception detection: ‘this looks different than normal.’
Decisions AI can execute conditionally
- Low-risk actions with clear rules and a rollback path.
- Routing based on explicit fields and validated inputs.
- Notifications and reminders with clear ownership.
Required override points
- Before external communication is sent (unless it’s truly low-risk).
- Before money, access, or irreversible actions occur.
- When confidence is low or inputs are incomplete.
Common mistakes
Common mistakes
- Letting “automation” own decisions because it’s faster.
- Hiding the override path so people can’t correct the system quickly.
- Treating drafts as final and removing human review prematurely.
What this does NOT answer
- Which specific model/tool to use.
- How to automate a moral decision (don’t).
Optional next step
If you want help drawing the boundary for a specific workflow, that’s what the AI Automation Audit is for.