Configure escalation and human support
Define real support channels, handoff criteria, pre-handoff questions, operator step-in, satisfaction, and ticketing.
Escalation determines when and how the customer-facing Agent involves a human. Current settings include ordered support channels, a human-handoff switch, include and exclude criteria, any/all matching, pre-handoff questions, operator step-in, satisfaction collection, and administrator-only ticketing integrations.
This guide focuses on configuration. Pair it with the Inbox takeover guide for the live operator workflow and with team access guidance so the customer promise is backed by staffed, authorized people.
How this fits into Humind
Inbox is the operational record of customer conversations. It combines message history, shopper and product context, internal collaboration, human-handoff state, tags, and tickets. Customer replies and internal notes have deliberately different visibility.
A consistent team routine matters more than any individual filter. Agree when to take over, when to leave an internal note, when to create a ticket, and when to return a conversation to the AI so ownership remains clear to every operator.
Before you start
Access: Agent configuration write access is required. Ticketing provider credentials are restricted to Humind administrators in the current interface. Operators need separate Inbox and handoff permissions.
- Approve staffed support channels and current contact destinations.
- Define scenarios that must reach a human and scenarios that must not.
- Prepare authorized operators, availability expectations, and queue ownership.
Step-by-step workflow
Build the support-channel list
Open AI Agent, then Escalation. Add only support channels the company currently operates, enter the correct destination, enable them, and drag them into the desired customer-facing order. Use the preview to review the resulting choices.
Disable or remove a channel when it is no longer monitored. A visible contact option should always lead to a real team or system.
Enable and scope human handoff
Turn on human handoff. Add include criteria in natural language for cases that should be queued, and exclude criteria for cases that should remain with the Agent. When there are multiple include criteria, choose whether any or all are required.
Keep criteria mutually understandable and testable. If an exclude rule overrides a common include case, document the intended result and test both paths.
Collect useful pre-handoff information
Add questions the Agent should ask before entering the queue. Focus on information an operator needs to act, such as the subject of the request, without collecting unnecessary personal or sensitive details.
Collected answers appear in the Inbox conversation context. Ask operators whether the questions reduce repeated work and remove fields that do not change the response.
Choose operator and satisfaction settings
Operator step-in allows an authorized teammate to reply to an AI-active conversation and take it over. Disable it for a strict queue-only model. Configure AI satisfaction collection separately; human satisfaction is available only when handoff is enabled.
Operator profile name, image, status, and availability should be reviewed by each teammate because they affect the live support experience and operational expectations.
Review ticketing and run the full test
Administrators can enable Humind ticketing or configure available Gorgias and Zendesk credentials in the Escalation area. Confirm provider ownership and a successful test before making it part of the process.
Test an included scenario, an excluded scenario, pre-handoff answers, the queue, operator reply, internal note, hand-back to AI, satisfaction, and ticket creation if enabled. Record the result from both shopper and operator views.
Permissions and important caveats
- The general Integrations marketplace can label Gorgias and Zendesk as early access even though administrator credential controls exist in Escalation. Verify actual provider readiness with Humind.
- Operator step-in policy is enforced; a forbidden AI-active takeover returns an error.
- Human satisfaction is unavailable when human handoff is disabled.
- Pre-handoff questions should follow the company's privacy policy and collect only necessary information.
Verify the result
Use this checklist before considering the work complete:
- Every enabled channel is accurate, staffed, and ordered intentionally.
- Include, exclude, and match-mode tests produce the expected queue behavior.
- Operators have access, availability guidance, and a documented takeover or hand-back routine.
- Ticketing and satisfaction paths are tested only when actually enabled.
Troubleshooting
Too many conversations enter the queue
Review broad include criteria, missing exclude criteria, and any/all matching. Narrow one condition at a time and rerun matching and non-matching test pairs.
A valid request never escalates
Confirm handoff is enabled, the active Agent uses the saved criteria, the request is not excluded, and support channels are enabled. Start a new shopper test after changes.
Operators cannot take over
Check Inbox and Inbox handoff permissions, operator availability, thread state, and the operator step-in setting. A strict policy can intentionally block replies to AI-active threads.