Escalations

Catch risky or important messages and route them to your team — notify staff, pause the AI, or hand over the conversation.

Escalations#

Some messages shouldn't be left to the AI alone — an angry customer, a legal question, a big order, a VIP. Escalations lets you set rules that watch incoming messages, comments, reviews, and emails, then take the right action: notify your team, pause the AI, hand the conversation to a human, or prepare a careful reply for approval.

Find it under Agent Studio → Escalations.

How It Works#

When a new message, comment, review, or email comes in, your rules decide whether it needs a person.

Code
New message / comment / review / email
            |
    Matches a review rule?
            |
   ┌────────┴─────────┐
  Yes                 No
   |                   |
Take action:      Agent replies
- Notify team       as normal
- Pause AI
- Hand over
- Draft a safe reply

Flagged work shows up in Activity (use filters like Needs reply or Pending approval), and your team can be notified on the channels you've set up under Settings → Team Channels.

Managing Rules#

Create and manage the conditions that route work to humans:

  • Source — all channels, or just Messages, Comments, Reviews, or Email
  • Condition — what to watch for (keywords, tone, or the type of request)
  • Action — notify staff, pause the AI, hand over the conversation, or prepare a draft reply
  • History — which rules fired and when

AlonChat can also surface rule suggestions from recent conversations (patterns that look like they should be escalated). Review suggestions and accept the ones that fit your business — often from Conversation Learning or related review surfaces in the product.

What These Rules Are (and Aren't)#

Escalations decides what happens when something notable occurs. It is not, by itself, a support ticket. Depending on your rules, a flagged item can simply notify you, pause the AI for one conversation, or — if you use it — create a Help Desk ticket for your team to own.

Best Practices#

  1. Start with the obvious risks — complaints, refunds, "speak to a human", legal/medical questions.
  2. Use suggestions when available to catch patterns you didn't think to add.
  3. Pair notify + pause for sensitive topics so the AI doesn't keep replying while a human steps in.
  4. Review rule history regularly to tune rules that fire too often or not enough.

Related Pages#