Skip to content
AskVault Docs
Try Free →

The escalate_to_human skill

Written by , Founder, AskVault · Reviewed by Aashiq
Last updated: · 5 min read

What the skill does

The escalate_to_human skill is the safety net under every AI customer-support deployment. The AI handles the 60 to 70% of questions that have a knowledge-base answer. When something falls outside that scope or the customer just wants a human, this skill takes over.

The handoff:

  1. The bot detects an escalation trigger.
  2. The bot tells the customer "Let me get a human to help". Stops auto-responding to the conversation.
  3. The conversation appears in your Live Chat inbox with full context: prior turns, retrieved sources, the trigger that fired.
  4. An agent picks it up. They reply in the inbox. The customer receives the reply in their original channel (widget, WhatsApp, Slack, etc.).

End-to-end handoff time: under 30 seconds for explicit "agent" requests. Up to 2 minutes for auto-detected frustration if the agent isn't actively watching the inbox.

When the skill fires

Three trigger types, configurable under AI Agents > Skills > escalate_to_human > Triggers:

  1. Explicit request. Customer says "human", "agent", "representative", "person", or similar. The trigger phrase list is editable.
  2. Frustration signal. The sentiment_router skill flags the conversation as angry or frustrated. Or the customer has rephrased the same question 3+ times without getting a satisfying answer.
  3. Low-confidence chain. The bot has answered with confidence: "low" for 2+ consecutive turns. The retrieval keeps coming back empty.

You can disable any trigger type. Some teams want only explicit requests, never automatic escalation. Others want the auto-detection to be aggressive.

Routing the escalation

Where the conversation lands depends on your routing rules under Automation Rules > Escalation Routing:

  • Default routing. Lands in Live Chat > Conversations with no specific agent assignment. Whoever picks it up first owns it.
  • Round-robin by agent. Distribute escalations evenly across your support team.
  • Skill-based routing. Route escalations with tags like billing to your billing team, technical to engineering, etc. Tags are auto-assigned by the sentiment_router skill or by your automation rules.
  • Time-of-day routing. Office hours go to your in-house team; after hours go to a third-party answering service or a specific on-call rotation.

For Slack and Discord channels, escalation can ping a specific user or role (@on-call-support) instead of routing to the Live Chat inbox. Configurable per channel.

What the agent sees

When an escalated conversation arrives, the agent sees:

  • Full conversation history. Every customer message and every bot reply, with timestamps.
  • The trigger that fired. "Customer said: 'I want to talk to a human'" or "Sentiment dropped to angry after 3 turns".
  • Retrieved sources. Which knowledge-base chunks the bot used for its replies. Helps the agent see what context the bot had.
  • Customer profile. If the conversation was identity-verified, the agent sees the customer's name, email, plan tier, and any custom fields you've configured.

The agent replies inline. The customer keeps chatting in their original channel and sees the agent's reply naturally.

Resuming the bot

After the agent finishes, two options:

  1. Resolve the conversation. Marks it closed. If the customer comes back later with a new question, the bot answers fresh.
  2. Pass back to the bot. The bot resumes auto-responding from the next message. Useful when the agent answered one specific question but the customer might have more.

Configurable per-conversation. Most teams default to resolve, since most escalations are about something the bot won't be able to answer either.

Customizing the handoff message

Under AI Agents > Skills > escalate_to_human > Message, customize what the bot says at handoff. Default:

Let me connect you with a human. They'll be with you in a moment.

For an after-hours scenario:

It's outside our business hours (Mon-Fri, 9am-6pm IST). I've flagged your conversation and a human will respond first thing tomorrow morning. Sorry for the wait.

For a slow-channel scenario (email):

I'll have a human look at this and get back to you within one business day. We aim to respond within 4 hours during business hours.

Match the message to the realistic response time of your team.

Per-channel escalation

The skill works on all 13 channels but the UX varies:

  • Website widget. Smooth real-time handoff. Customer sees a typing indicator from the human.
  • WhatsApp, Telegram, Discord. Conversation continues in the customer's channel. Human types from the inbox.
  • Email. Conversation switches to draft-for-review mode. Human writes the next reply.
  • Voice. Bot transfers the call to a configured handoff number. Hold music for a few seconds, then human picks up.
  • SMS. Human replies via the Twilio number. Customer doesn't notice the transition.

Limits and pricing

The skill is on Growth and above. Growth+

Each escalated conversation counts as 1 query (the trigger detection step). The human's responses don't count against AskVault's query quota; they're not AI-generated.

Common pitfalls

Bot escalates too aggressively. Auto-triggers are firing on legitimate questions. Loosen the sentiment threshold under Triggers > Sentiment > Threshold or disable auto-trigger entirely.

Bot doesn't escalate when customer asks for human. Check the trigger phrase list. "Human" alone might not be in the list if you edited it. Add "human", "agent", "person", "representative", "real person".

Escalations pile up unattended. The Live Chat inbox isn't being watched. Enable Slack notifications under Settings > Notifications > Slack to ping the support channel when a new escalation lands.

Agent replies don't reach customer. Most likely the agent is replying to the wrong conversation tab. Check that the agent is in the inbox view, not the playground.

FAQ

Can the bot escalate to email instead of live chat?

Yes. Configure escalation to email under Automation Rules > Escalation Routing > Send to Email. The conversation context is forwarded as an email; replies happen async.

Does the customer know they've been transferred to a human?

By default, the bot says so explicitly ("Let me connect you with a human"). If you'd rather hide the transition (single-persona support feel), customize the handoff message to omit that disclosure.

Can I have a hybrid where the agent reviews the bot's draft before sending?

Yes. That's draft-for-review mode, the default for the Email Assistant channel. The bot drafts replies, the agent reviews and sends. Useful for high-stakes channels where you don't want auto-responses.

How fast does the customer get a human?

Depends on whether an agent is actively watching the inbox. With a dedicated agent at the inbox, under 30 seconds typical. With a passive watch and Slack notification routing, 2 to 5 minutes typical.

Does the bot pause completely after escalation?

Yes, for that conversation. The bot stops auto-responding once the skill fires. New conversations from other customers still get bot responses.

Was this page helpful?