Answering
The handoff is judged on what the person receives, not on how fast it fired.
Every assistant can escalate. The difference between one that helps a team and one that adds a step is entirely in the payload.
- Private launchOnboarding selected teams now
- Answers from your pages onlyIt refuses when your content does not cover it
What it does
Handoff moves a conversation from the assistant to a person, carrying the full transcript, the retrieval trail, the trigger that fired and the page the visitor was on.
Why it matters
The common implementation escalates to a contact form. The visitor, who already explained their situation, is asked to explain it again, and support receives a message with no history. The first reply is then a discovery question the visitor has answered twice.
Carrying the retrieval trail is the part most implementations skip and the part that makes a transcript useful afterwards. Without it you know the assistant failed but not why, so the log tells you nothing about which page to fix.
How it works
Triggers are written rules
Explicit request, repeated failure on one thread, a topic marked always human, or low confidence on a high stakes question. Rules you can read and audit, not a judgement made fresh each time.
The assistant says so plainly
It states that it is handing over and why. A silent transfer reads as the widget breaking.
The payload is assembled
Full transcript including failed attempts, the pages searched and the passages returned, the trigger that fired, the page and time on it, and any contact detail already given.
Delivery
The conversation is routed to your inbox or channel with that payload attached.
The gap is logged
If a content fix would have prevented the escalation, it is flagged. That flag is what turns a support log into a product input.
The failure mode that looks like success
An escalation rate that falls month on month is usually reported as improvement. Sometimes it is. Often it means the assistant has become more willing to guess.
Both a good assistant and an overconfident one show low escalation. The number that separates them is refusal rate, and the two should move together. Escalations falling while refusals also fall is the pattern worth investigating, because it means questions that used to reach a person are now being answered by something that does not know it is wrong.
The other misleading direction is an escalation rate that rises after a content change. That often means the assistant is correctly declining to improvise on a topic your new page half covers, which is the system working.
Read the escalations rather than counting them. Twenty a week that a content fix would have prevented is a content problem. Twenty a week that genuinely needed a person is a staffing decision. The count alone cannot tell you which you have.
What is built and what is planned
Live handoff, with a person joining the conversation in real time, is planned for launch on the Scale plan. Today the accurate description is that a conversation can be escalated with its payload, not that someone is waiting.
Where the data goes
Transcripts are stored so they can be handed over and reviewed later. Retention periods will be published before paid rollout. Security and data boundaries.
Where this comes up
Questions
Signal
Is this the capability you need?
Access is by waitlist, demo request or public-content pilot. Tell us what your site has to answer and we will say honestly whether this covers it.