Working with Human Moderators (Escalation & Hand-off)

Your eChat agent is built to work with your team, not in front of it. In eWebinar it acts as an AI moderator: a member of your moderator team in the same chat as your humans. Most of the time nothing is "escalated" at all — the AI and your people simply share the work.

A real hand-off, where the conversation is passed to a person, is the exception. It only happens when your agent uses Bring in a person (Bring in a moderator for eWebinar) and your settings allow visitors to reach a person. This article covers how the AI works with your moderators day to day in eWebinar, the wait time that gives them the first word, flagging, notes, stepping back when someone jumps in, when a real hand-off happens, and what to check when something looks off.

How the AI works with your moderators

What it is

In eWebinar, the AI moderator behaves like a colleague on shift with your team. It takes the routine questions, leaves anything that needs a person clearly marked, and gets out of the way when a human is handling the chat.

How it works

  • It knows the team. eWebinar keeps the AI up to date on which of your moderators are around, so it only mentions a colleague when it knows they're available.
  • It gives humans the first word. With a wait time set in eWebinar, your moderators are notified when a question comes in, and the AI only answers if nobody replies in time. See the wait time below.
  • It marks work for humans without walking away. The AI can flag a specific attendee message as Needs reply and leave a private note explaining why, then keep helping the attendee. Flagging can also email your moderators so they see the note.
  • It steps back when a human steps in. When a moderator replies, starts typing, or leaves a private note, the AI adjusts rather than talking over them.
  • It hands off only when you've allowed it. A person is brought in only with Bring in a person (Bring in a moderator), on webinars where eWebinar allows it.

How much of this the AI does depends on Escalation on the agent's Agent screen. With Agent only it works on its own. Leave a note adds flagging and notes. Bring in a person adds real hand-offs on top. See choosing a mode.

Best practices

Set your moderators' availability in eWebinar and keep it accurate. It's what the AI uses to decide whether to wait for a person and how to talk about your team.


The wait time: giving your team the first word

What it is

Each webinar's AI moderator settings in eWebinar include AI answers if a human doesn't answer in … sec, in the AI moderator's Behavior section of the webinar's Moderators settings. This is an eWebinar setting for each webinar, not something you set in eChat. It only affects conversations in eWebinar.

How it works

  • Your team gets the first word. When an attendee asks something, your moderators are notified and the AI holds back. If a moderator replies or starts typing, the AI lets them handle it.
  • No question goes unanswered. If nobody replies in time, the AI answers, covering everything the attendee sent in the meantime. Your moderators see a countdown in the eWebinar inbox.
  • No waiting when there's nobody to wait for. If no moderator is available, or the attendee is already chatting with the AI, it answers right away.
  • Room for your moderators. After one of your moderators replies, the AI gives them a moment to answer the attendee's next message too.
  • No credits while it waits. If a human answers first, the AI doesn't reply to that message.

Set the wait anywhere from 0 (the AI answers right away) to 15 minutes. It applies on webinars where attendees can request escalation; for Agent only and Leave a note agents, the AI answers right away. With a wait above 0, moderators are notified about each new question. With 0, they're notified only about hand-offs and messages flagged as Needs reply.

Best practices

  • If your team actively moderates live sessions, keep a short wait (the 2-minute default is a good start) so people answer first and the AI fills the gaps.
  • For automated sessions nobody watches, use 0. The AI answers at once, and your team only hears about flagged or handed-off conversations.
  • If you start typing and change your mind, send the attendee something anyway. Typing cancels the AI's answer to that message.

Flagging messages as Needs reply

What it is

When a person should follow up on something, the AI marks that attendee's message as Needs reply. It does this for things it couldn't fully resolve, account-specific issues, bug reports, or requests that need a person after the webinar. This is not a hand-off. The AI keeps talking with the attendee, and the flag just makes sure a human sees the message. Your moderators may also get an email when the agent flags a message for them.

How it works

  • Only your moderators see flags.
  • In the eWebinar inbox, the Needs reply filter lists conversations with flagged messages. On desktop it sits next to Unread. On phones it's in the mailbox menu under Inbox. It appears where the AI moderator is an eChat agent.
  • Flagging doesn't hand the conversation over and doesn't switch the AI off.
  • A flag usually comes with a private note explaining why.
  • Other things raise a Needs reply flag too: a real hand-off, an attendee asking for a person, and your account running out of credits (the attendee's newest message is flagged so your team can cover it).
  • When a moderator replies to the attendee, that usually settles a flag the AI raised. Flags from a hand-off or an attendee's request for a person stay until someone on your team deals with them.
  • Flagging is available with Leave a note and Bring in a person.

Best practices

Use Needs reply, not Unread, as your team's to-do list. When the AI replies, the conversation counts as read, so flagged conversations won't necessarily show as unread.


Leaving private notes

What it is

The AI can leave a private note on the conversation for your moderators. The note is colleague-to-colleague context that only your team sees.

How it works

  • Notes appear to moderators as yellow Private note from <agent> bubbles (the same style as eWebinar moderator notes). In eChat Chat Logs view-only replay, those yellow notes are included so you can review what the AI told your team.
  • The AI leaves a moderator note only on turns that matter to your team: a real hand-off, an offer to bring a person in, or a reply that flags a message as Needs reply.
  • Notes are written in the attendee's language. Every real hand-off comes with a note saying what the attendee actually needs.
  • With Leave a note, the AI may briefly tell the attendee it has left a note so your team can give a more accurate answer later.
  • Poll, quiz, and question-card answers are the attendee's own answers, not notes. The AI can respond to them in chat, or stay quiet if your Behavioral instructions say so. See how the AI answers.

Best practices

Treat the AI's notes as a brief. If the same reasons keep showing up, answer them once in Q&A so the AI can handle them itself next time.


Stepping back when a human jumps in

What it is

The AI stays in the conversation even after one of your moderators joins, because it's a peer rather than a bot that gets switched off. But it takes its cue from the humans.

How it works

  • You reply or start typing. The AI's pending answer to that message is cancelled, and it gives you the first word on the attendee's next message. See the wait time.
  • You leave a private note. The AI reads it straight away as a note from a colleague, not as a question from the attendee. It replies to the attendee only if the note makes that useful. Otherwise it stays silent.
  • You hand the conversation back. When a moderator turns the AI back on for a conversation, the AI answers the attendee's waiting message right away, with no wait, or stays quiet if there's nothing to answer.
  • It picks up after a colleague. If the AI answers after a human was talking with the attendee, it says so naturally.

Best practices

If you want to handle a conversation entirely yourself, turn the AI off for that conversation in eWebinar, and turn it back on when you're done. Use private notes to steer the AI rather than typing instructions where the attendee can see them.


Choosing an escalation mode

What it is

Escalation is three fixed steps in order: whether the agent involves a person at all, then when, then how. Step 1 lives under Escalation on the agent's Agent screen.

Escalation settings in eChat showing the Chatbot only, Leave a note and Human hand-off modes, with the Answer is uncertain, Frustrated language and Human request follow-up toggles

How it works

Agent only. The AI is the only help available. If a visitor asks for a person, it keeps helping them itself. Choosing Agent only hides the when and how steps.

Leave a note. No live hand-off. The AI can flag messages as Needs reply, leave private notes for your team, and may tell the visitor it has done so. It stays the visitor's helper for the whole chat without offering to bring someone in.

Bring in a person (Bring in a moderator on eWebinar agents). Everything Leave a note does, plus the AI can hand the conversation to a person, based on the when triggers you turn on and what eWebinar allows for each webinar. This is the only mode where a person can be brought into the conversation. See when a real hand-off happens.

For eWebinar agents, how is a statement — your eWebinar moderators.

Best practices

  • Pick Agent only only if nobody on your team will ever look at these chats.
  • Leave a note suits automated sessions. Your team gets a clear list of what needs a person without promising anyone live help.
  • Choose Bring in a person when people really are around and can pick up a conversation.

When a real hand-off happens

What it is

A hand-off passes the conversation to your human moderators. It only happens in Bring in a person mode, and only when the webinar's eWebinar settings allow attendees to reach a person.

How it works

eWebinar decides whether a hand-off is possible. In each webinar's AI moderator settings:

  • Attendees can request escalation must be on. eWebinar only lets you turn it on when at least one human moderator is assigned to the webinar and first-chat-message notifications are on (Slack, email, or SMS). See prerequisites for human hand-off.
  • Only allow escalations when a human moderator is available is optional. When it's on, neither the attendee nor the AI can start a hand-off unless an assigned moderator is available at that moment.

If escalation is off, or only allowed while someone is available and nobody is, the AI won't offer or start a hand-off. It keeps helping the attendee itself. If escalation is allowed but nobody is available right now, the AI can still pass the conversation to your team, but it tells the attendee nobody is available live and that someone will follow up later.

What happens during a hand-off.

  • The AI tells the attendee that a person is being brought in, without naming who or promising when they will arrive.
  • Every moderator assigned to the webinar is notified through your chat notification settings.
  • The attendee's message is flagged as Needs reply, and the AI leaves a note explaining what they need.
  • The AI stays in the conversation. Its reply is the holding message, and it keeps helping while your team gets there.
  • If nobody picks the conversation up in time, the AI follows up with the attendee, making clear the team has already been told, without promising a new connection.
  • Once a human replies, that request is done. If the attendee later needs a person for something new, the AI treats it as a fresh request.

Best practices

Keep the Needs reply filter open during live sessions so hand-offs don't sit waiting. The AI keeps the attendee informed, but only a person can finish a hand-off.


What it is

In Bring in a person mode, the AI asks the attendee before handing off rather than paging your team on its own judgment.

How it works

When the AI hits a reason to involve a person (it isn't confident about an on-topic answer, the attendee sounds frustrated, reports a problem, or needs something tied to their own account), it helps as far as it can and then asks whether they'd like a person to help. The attendee accepts by typing yes (or otherwise clearly agreeing). Nothing is handed off, and no one is flagged, until they accept. If they decline or move on, the AI keeps helping.

The one exception is an explicit request. If the attendee asks for a person, that request is their consent, so the AI can hand off without asking again, as long as the Human request trigger is on.

Best practices

Match the mode to your real staffing rather than relying on the ask. If nobody will see a hand-off, choose Leave a note instead.


When: Human follow-up triggers

What it is

With Bring in a person selected, the agent's Agent screen shows the when cards that decide which situations can lead to involving a person: Answer is uncertain, Frustrated language, and Human request.

How it works

  • Answer is uncertain: the AI can offer a person when it can't answer an on-topic question confidently.
  • Frustrated language: the AI can offer a person when the attendee sounds blocked, upset, or like the answer didn't help. This leads to an offer, not an automatic hand-off.
  • Human request: the AI hands off when the attendee explicitly asks for a person or follow-up.

All three are on by default. Turning one off means that situation no longer leads to an offer or a hand-off. Turning off Human request also means an explicit request is no longer handed off automatically. Reported problems and account-specific questions have no toggle of their own — with hand-off available, they can always lead to an offer, and the attendee still has to accept.

These cards also depend on eWebinar. If escalation is turned off for every linked webinar, the cards are disabled and a notice explains why. Each webinar's own setting still applies to its conversations.

Best practices

Leave Human request on unless you truly have nobody to hand off to. If moderators get pulled in for questions the AI should know, close the training gap before turning off Answer is uncertain.


Troubleshooting

The AI didn't answer right away. That's usually the wait time working: a moderator was available, so the AI gave them the first word. It also gives a moderator a moment after their own reply in that conversation. Check the webinar's AI moderator settings in eWebinar, or set the wait to 0 for sessions nobody watches.

The AI didn't answer a message. A moderator probably replied, or started typing, during the wait. Either one cancels the AI's answer to that message. If the moderator abandoned their draft, reply to the attendee directly.

The AI keeps replying after I jumped in. That's expected: the AI is a peer and stays in the conversation. After your reply, it waits before answering the attendee's next message. To take a conversation over completely, turn the AI off for that conversation in eWebinar.

An attendee asked for a person but nobody was brought in. Check that the mode is Bring in a person, that Human request is on, and that the webinar allows attendees to request escalation. If the webinar only allows escalation while a moderator is available, nobody may have been available.

I can't turn on attendee escalation in eWebinar. The webinar needs at least one assigned human moderator and first-chat-message notifications (Slack, email, or SMS).

Moderators aren't notified about new questions. With a wait of 0, eWebinar only notifies moderators when a conversation is handed off or the AI flags a message. Set a wait above 0 to notify them about every new question.

The Needs reply list is empty. With Agent only, the AI doesn't flag messages. Otherwise, the AI may simply have handled everything itself. Check Chat Logs to review its answers.

The AI stayed silent on a poll or card answer. Check Behavioral instructions — if they tell the agent to stay silent or reply only for other answers on notify-moderators cards, it posts nothing. Typed messages are still answered.

The AI stopped answering and messages are being flagged. Your account has probably run out of credits. The AI goes quiet and the attendee's newest message is flagged for your team. See Plans, Credits & Usage.

The same questions keep getting flagged. Add answers to Q&A. Conversations themselves aren't training data.