Assignment & escalation

Assignment & escalation

A conversation has three lives: bot-handled, single-agent, and multi-agent. Knowing how each transition works lets you design a queue your team won’t drown in.

In this guide:

  • When the bot escalates
  • Auto-escalation triggers
  • Manual escalation by the user
  • Assignment patterns
  • Reassignment

When the bot escalates

The bot hands off in three situations:

  1. Confidence too low. The bot’s response confidence dropped below the escalation threshold you configured. Tune that threshold to find your sweet spot — too low and the bot sticks too long; too high and humans get spammed.
  2. Explicit user request. Phrases like “talk to a human”, “agent please”, “this isn’t helping” trigger immediate escalation. The trigger phrases are configurable in Behavior.
  3. Action failure. A required AI Action failed and the bot couldn’t recover. Better to hand off than guess.

In all three cases, the conversation moves from bot-only to Unassigned in the agent dashboard.

Auto-escalation triggers

Beyond the basic “low confidence” trigger, configure additional rules in Behavior → Escalation triggers:

  • Sentiment — escalate if the user shows frustration (configurable thresholds).
  • VIP customer — verified-identity users on a specific tier get fast-tracked.
  • Specific topics — keyword-based (“refund”, “complaint”, “cancel my account”).
  • Conversation length — escalate after N user messages without resolution.

Manual escalation by the user

Even without trigger phrases, users can manually escalate:

  • A “Talk to a human” button in the widget (if enabled in Appearance).
  • A reply containing the words you’ve configured (“agent”, “human”, “support”).

Assignment patterns

When a conversation lands in Unassigned, three common patterns:

  1. Free-for-all — any agent can pick from Unassigned.
  2. Round-robin — auto-assign to the next available agent. Configure in Settings → Routing.
  3. Skill-based — route by tag, channel, or chatbot. Conversations from #refunds go to the refunds team.

For larger teams, mix: round-robin within skill groups.

Step: Reassignment

Anyone with conversations.assign permission can reassign:

  1. Open the conversation.
  2. Click the assignee dropdown.
  3. Pick a teammate (or Unassign to put it back in the queue).
  4. Add a note (“Reassigning — they own this account”).

Assignment dropdown Screenshot: The assign-to-teammate dropdown in a conversation.

The reassigned agent gets a notification.

SLAs and timers

For plans that support it:

  • Each conversation has an SLA timer (first-response and full-resolution).
  • Conversations approaching their SLA show a yellow indicator; past-SLA show red.
  • SLA misses are tracked in Performance analytics.

Configure SLAs in Settings → SLAs.

Pause / snooze

For conversations waiting on the customer:

  • Snooze for N hours/days. The conversation drops out of your active queue and re-appears when the snooze expires (or sooner if the customer replies).
  • This keeps your My conversations queue clean without losing track.

Tips

  • Pick a single assignment pattern. Mixed patterns confuse agents.
  • Document tag conventions. “#vip” should mean the same thing org-wide.
  • Tune the escalation threshold weekly for the first month — your data will tell you when the bot is being too aggressive.

Troubleshooting

  • Conversations sit in Unassigned forever. No one’s watching the queue. Set up Slack notifications via the Slack channel integration.
  • Agents complain about too many escalations. Lower the escalation threshold and review which queries shouldn’t have escalated.
  • Round-robin assigns to inactive agents. Only assign to agents who are “online” — toggle in Settings → Routing.

What’s next