liveBook · Manning

Handoff

As explained in Effective Conversational AI

Escalation is the transfer of a chatbot interaction to a human agent or other additional handling when the bot cannot complete the request or the user requests assistance.

A user may reach the end of a claim lookup and still ask for a representative. Handoff transfers the interaction from a chatbot to a human agent when the user requests help or the bot cannot complete the task. The transfer should help the agent continue the work, not make the user start over. section 8.1.2 section 12.0

Choose the next step according to what went wrong:

  • Retry when an input is incomplete, but do not repeat the same confusing question indefinitely. Users may request an agent before the bot reaches its retry limit. section 8.2.5 section 11.1.2
  • Offer another route when one is available, such as looking up an account by phone number after an account-number lookup fails. section 8.2.5
  • Hand off when retries cannot help, such as during a backend outage that requires manual processing. section 8.2.5

A transfer need not mean the bot failed: some business processes intentionally send users to humans. By contrast, escalation after repeated misunderstanding can signal a problem. Track these outcomes separately, then inspect where escalations occur. A spike at a particular question may point to wording or response formats that need improvement. section 3.3.2 section 3.4.1

A transfer that preserves progress

After supplying claim details and receiving a payment result, a user says, “representative.” A blind transfer would force the user to repeat those details; sending only the full transcript would make the agent search while the user waits. A targeted conversation summary lets the agent see what was requested and accomplished and pick up promptly. section 12.1.1 section 12.0

Listing 12.1 Example conversation between user and AI
AI: Welcome to the automated assistant, how can I help?
User: Claim status
AI: Ok, claims. What's your Tax ID?
User: 123456789
AI: Thank you. What's the member's eight-digit Member ID?
User: 87654321
AI: Ok, and what's the member's date of birth?
User: 12/31/2000
AI: Is the member's name John Doe?
User: Yes
AI: Great! What's the date of service for the claim
User: "February first 2024
AI: I found your claim #111222333 from February 1,  #1
2024 for the member #87654321 and provider #123456789.  #1
It was paid on May 23, 2024 for $201.83.  #1
User: representative  #2

The claim-status dialogue shows how much information an agent would need to recover if the user requests a representative after several turns. section 12.1.1

How Effective Conversational AI covers it

How deep it goes
Explains when to retry, offer another path, or transfer, and why a targeted summary improves handoff; does not detail transfer implementation.
Examples
Uses a claim-status dialogue ending in an agent request, an outage requiring manual processing, and a lookup with an alternative route.
What you can do after
Practice/perform handoffs that avoid needless repetition, and diagnose escalations by conversation outcome and last step.
How it connects
Links handoff to retries, self-service, conversation outcomes, containment, and conversational summarization.

Also explained in

Compare how 2 books explain handoff →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Effective Conversational AI. The section links point to the parts of the book it draws on.