Handover Modes Reference

Each Agent Stack has one handover mode controlling whether and how the AI hands off conversations to humans. The mode is set on the Handover → Triggers sub-tab (off/on switch) and refined on the Delivery sub-tab (which mode and which team).

The three modes

• never — The AI never hands off — Stays in the stack until resolved or the customer leaves
• by_description — When trigger conditions match (frustration, legal threat, escalation request, stall, “ask for a human”) — A team derived from conversation context (channel, topic, language)
• explicit_team — When trigger conditions match — A single team you pick from a dropdown

When to use each

never

Use never only for stacks that handle a single, narrow, closed task where there’s no meaningful human fallback — a sandbox stack, an internal-only stack, a survey collector. For real customer-facing stacks, leave handover on. Customers will eventually need a human, and never means they leave instead.

When the Allow handover to humans switch on the Triggers tab is off, the Prerequisites and Delivery tabs are greyed out — they only apply when handover is enabled.

by_description

The default. The AI decides which team picks up based on conversation context — what topic the conversation is about, what channel it came in on, what language it’s in. Pair this with a Handover instructions text in the Delivery tab to steer the AI’s choice.

This mode is most useful when you have multiple specialist teams (Billing, Tier 2 Support, Returns) and the topic of the handed-off conversation should determine who takes it.

explicit_team

Every handover goes to one specific team. Set the Handover team field on the Delivery tab to pick which team. Use this when:

• Your stack only covers one functional area, so one team always picks up.
• Your team structure is small and routing-by-description would be overkill.
• You want predictable, auditable routing for compliance reasons.

Triggers (when handover fires)

Regardless of mode, handover triggers when any of these are detected:

• The customer explicitly asks for a human (“can I talk to a person?”).
• The sentiment-and-risk infrastructure agent flags the conversation as risky (frustration, legal threat, churn risk, escalation request).
• The orchestrator detects a stall — the conversation has stopped progressing.
• A prerequisite check completes (for stacks where the AI’s job is to gather and hand off).

Required information

Before handing off, the AI tries to collect everything in the Prerequisites sub-tab. Each prerequisite has an always or when condition, and the AI asks the customer for any missing fields before triggering the handoff. If the customer refuses or stalls, the orchestrator can still trigger handover under risk logic.

Delivery options for outside opening hours

When Check opening hours is on (Delivery tab) and the team is closed:

• Offer email follow-up — on — AI asks for an email and tells the customer a reply will come during hours
• Offer close and return — off — AI cleanly closes the conversation with instructions to reopen during hours
• Ask confirmation — off — AI confirms the customer still wants to wait, before queueing them

Multiple options can be on at once — the AI presents them as choices to the customer.

Handover instructions

The free-form Handover instructions text on the Delivery tab is appended to the AI’s prompt when it’s deciding how to hand off. Use it for:

• Tone for the handoff message (“apologize for the wait if the conversation lasted more than 5 turns”).
• Information to always include in the handoff pack (“mention loyalty status if it’s Gold or Platinum”).
• Things to never reveal in the handoff message (“don’t tell the customer which team is picking up”).

See also

• Handover to humans
• Configure handover
• Recipe: graceful handover to a team