Skip to content
Clienta.ai

Human Handoff

Human Handoff

Section titled “Human Handoff”

Human Handoff is the bridge between your AI bot and your human support team. When the bot is uncertain or a customer needs personal attention, the conversation is automatically escalated to an available agent — smoothly, without the user having to repeat themselves.

How It Works

Section titled “How It Works”

  1. User sends a message — The bot processes the message using the RAG pipeline and generates a confidence score.

  2. Confidence check — If the confidence score falls below the configured threshold, the bot triggers a handoff instead of guessing.

  3. Handoff notification — The conversation moves to Pending Handoff in the Conversations inbox. Agents are notified.

  4. Agent takes over — An available agent accepts the conversation and begins replying. The full conversation history is visible to the agent.

  5. Resolution — The agent resolves the issue and closes the conversation. The bot can handle new sessions from the same user in the future.

Smart Routing

Section titled “Smart Routing”

When a handoff occurs, the system routes the conversation to the most appropriate agent using these signals:

SignalDescription
AvailabilityRoutes to agents who are currently online and not at full capacity
WorkloadBalances conversation load across available agents to prevent any single agent from being overloaded
Sentimentv1.6.0 Routes conversations with detected negative sentiment to senior agents for priority handling

Agent Typing Indicator

Section titled “Agent Typing Indicator”
v1.6.8

When a human agent is composing a reply in the dashboard, the visitor sees a typing indicator in the widget — the same animation used for AI responses. This gives visitors real-time feedback that their agent is actively working on a response.

The typing indicator is automatically cleared when:

  • The agent sends their message
  • The conversation is resolved
  • The agent navigates away from the conversation

Confidence Indicator & Smart Escalation

Section titled “Confidence Indicator & Smart Escalation”
v1.6.5

The widget includes a confidence-aware escalation system that gives visitors real-time feedback when the AI is uncertain about its answer.

How It Works

Section titled “How It Works”

Each AI response carries an internal confidence score. When the score is low (indicating the bot may not have a reliable answer), one of two behaviors occurs depending on your handoff configuration:

ScenarioWhat the visitor sees
Low confidence + auto-handoff enabledA status indicator appears: “Connecting you to our team…” — the system automatically routes the conversation to a human agent.
Low confidence + manual handoff onlyA “Talk to our team” chip appears below the message. The visitor can click it to request a human agent.
High confidenceNo escalation UI is shown — the conversation continues normally with the AI.

Behavior Details

Section titled “Behavior Details”
  • The escalation indicator only appears on the most recent low-confidence message, not on older ones.
  • Once a human agent picks up the conversation (status changes to “assigned”), all escalation UI is hidden.
  • The “Talk to our team” chip disappears after the visitor clicks it once, preventing duplicate handoff requests.
  • Agent replies (from human agents) are never flagged for escalation, even if metadata is present.
  • All escalation text is fully localized across all 15 supported languages.
  • The escalation indicator uses role="status" and aria-live="polite" for screen-reader accessibility.

Priority & SLA

Section titled “Priority & SLA”

Priority Levels

Section titled “Priority Levels”

Conversations can be assigned a priority level, which affects queue position:

  • P1 — Urgent — Appears at the top of the queue regardless of arrival time
  • P2 — High — Elevated priority, surfaced above normal conversations
  • P3 — Normal — Standard queue order (default)
  • P4 — Low — Deferred; handled after higher-priority conversations

Priority can be set manually by an agent, or assigned automatically based on rules (e.g., VIP customer segments).

SLA Monitoring

Section titled “SLA Monitoring”

SLA (Service Level Agreement) targets define the maximum response time for each priority level. The system monitors all open handoff conversations against these targets:

  • SLA timer starts when the conversation enters Pending Handoff status
  • SLA at risk — Warning shown when 80% of the SLA time has elapsed
  • SLA breached — Alert triggered when the deadline is exceeded

Breach Alerts

Section titled “Breach Alerts”

When an SLA is breached:

  • The conversation is flagged with a Breached badge in the inbox
  • Admins receive an in-app notification
  • The breach is recorded in Analytics for reporting