Escalation Router Overview

The Escalation Router enables your AI agents to request human input when they encounter situations requiring approval, clarification, or human judgment. It provides a centralized system for managing all human-in-the-loop interactions across your organization.

Purpose

When AI agents process emails and execute tasks, they sometimes encounter situations that require human oversight:

• Decisions above their authority level (large purchases, policy exceptions)
• Ambiguous instructions needing clarification
• Compliance or legal concerns requiring expert review
• Deadline risks requiring stakeholder notification
• Requests for additional resources or access

The Escalation Router provides a streamlined workflow for these scenarios:

1. Agent creates escalation - The agent uses the request_human_input tool to pause and request human guidance
2. Email notification sent - An email is sent to the designated recipient. It comes from your dedicated escalation mailbox if you have one, and from your AI Chief of Staff's mailbox if you don't (see Delivery Modes)
3. Human responds - The recipient can respond via email reply or one-click web buttons
4. Agent resumes - The response is routed back to the waiting agent, which continues with the human's decision

Key Features

Unified Escalation Tool

All agents (both standard AI Agents and SME Agents) use the same request_human_input tool for human escalations. This provides consistent behavior and centralized visibility regardless of which agent type creates the escalation.

Multiple Request Types

The system supports various escalation scenarios:

Request Type	Description
Approval Required	Agent needs approval for an action
Clarification Needed	Instructions are ambiguous
Exception Request	Requesting an exception to policy
Compliance Concern	Legal or compliance issue detected
Deadline Risk	Risk of missing an important deadline
Resource Needed	Requesting access or resources
Information Request	General question requiring human input

Flexible Response Methods

Recipients can respond to escalations in two ways:

1. Email Reply - Reply to the escalation email starting with keywords like APPROVED, DENIED, or MORE INFO
2. Web Portal - Click Approve, Deny, or Request More Info buttons in the email to open a web form

Both methods route responses back to the waiting agent automatically.

Urgency Levels

Escalations support four urgency levels to help recipients prioritize their responses:

• Low - Non-urgent, can wait several days
• Normal - Standard priority (default)
• High - Needs attention soon
• Critical - Requires immediate attention

Auto-Escalation

For SME Agents with configured managers, you can set up automatic escalation to an alternate contact if the primary recipient doesn't respond within a specified timeframe.

Two Delivery Modes

The Escalation Router works whether or not you have a dedicated escalation mailbox. See Delivery Modes below.

Centralized Dashboard

All escalations are tracked in the Escalation Router management page, providing:

• Total escalation counts and resolution rates
• Pending escalations awaiting response
• Average response times
• Approval/denial rates
• Historical audit trail

How It Works

Agent Encounters Decision Point
              |
              v
    request_human_input tool
              |
              v
+-------------------------+
|  Escalation Created     |
|  (AgentEscalations)     |
+-------------------------+
              |
              v
+-------------------------+
|  Escalation Router      |
|  Sends Email (15s)      |
+-------------------------+
              |
              v
+-------------------------+
|  Manager Receives       |
|  Escalation Email       |
+-------------------------+
              |
      +-------+-------+
      v               v
  Email Reply    Web Portal
      |               |
      +-------+-------+
              |
              v
+-------------------------+
|  Response Processed     |
|  Notification Created   |
+-------------------------+
              |
              v
+-------------------------+
|  Agent Resumes with     |
|  Human's Decision       |
+-------------------------+

Delivery Modes

The Escalation Router has two delivery modes. Which one you are in is worked out automatically from your configuration, every time an escalation is sent. There is no setting to pick, and nothing to migrate when you move between them.

	Dedicated mailbox mode	Shared-sender mode
When it applies	You have completed the Escalation Router wizard and have an active escalation mailbox	Any other time (this is the default for a new tenant)
Sender address	Your configured escalation mailbox (e.g., escalations@company.com)	Your AI Chief of Staff's mailbox (see the sender order below)
"From" name	Your configured display name	"AI Escalations", so recipients see an escalation identity rather than an agent's name
Approve / Deny buttons	Yes	Yes, identical
Email replies	Accepted and routed back to the waiting agent	Accepted and routed back to the waiting agent, identical

The only thing that differs between the two modes is which address the escalation email is sent from, and the name shown next to it. Everything a recipient sees or does in response is the same either way.

How the sender address is chosen

Outermind picks the sending mailbox in this order, using the first one that is available:

1. Your escalation mailbox - an active monitored mailbox configured through the Escalation Router wizard
2. Your AI Chief of Staff's mailbox - the mailbox assigned to your AICOS agent
3. Any other active monitored mailbox in your tenant, most recently updated first
4. The Outermind system address - a last-resort sender used only when your tenant has no monitored mailbox at all

Steps 1 through 3 are all mailboxes in your own tenant, so a recipient can reply to the escalation email and the reply is routed back to the waiting agent. Step 4 is not a mailbox in your tenant, so escalation email sent that way carries a note telling the recipient that replies to the address are not monitored and asking them to use the buttons instead.

If a mailbox in the list turns out to have no Exchange mailbox behind it, Outermind moves to the next one automatically rather than failing the escalation.

Switching to a dedicated mailbox later

You can start in shared-sender mode and provision a dedicated escalation mailbox whenever you want. Complete the Configure Escalation Router wizard, and the next escalation is sent from the new mailbox. There is no migration step, no backfill, and no change to how recipients respond. Escalations already waiting for a reply are unaffected.

The reverse is equally automatic: if the escalation mailbox is later removed or reset, the router falls back to shared-sender mode and keeps working.

info

Shared-sender mode is always available as the fallback. If you do not want escalation email leaving from your AI Chief of Staff's address, provision a dedicated escalation mailbox: that is the way to change the sender.

Prerequisites

Before your agents can escalate, ensure you have:

1. Microsoft 365 Connection - A configured M365 connection in Capabilities > Connections > Microsoft 365
2. At least one monitored mailbox - Any active mailbox Outermind can send from. Most tenants already have one after setting up their AI Chief of Staff or their first monitored mailbox, so this usually needs no extra work.

A dedicated Microsoft 365 shared mailbox for escalations (e.g., escalations@company.com) is optional. Without one, the Escalation Router runs in shared-sender mode and escalations work normally.

The optional setup wizard that provisions a dedicated mailbox has one further prerequisite of its own: the Configure Safety Gateway wizard must be completed first.

Getting Started

The Escalation Router needs no setup to start working: a new tenant is in shared-sender mode from the beginning, and agents can escalate as soon as they exist.

If you want a dedicated escalation mailbox, run the optional setup wizard:

1. Navigate to Platform Setup
2. Open the Configure Escalation Router wizard
3. The wizard will guide you through selecting a shared mailbox, verifying access, and configuring settings

This wizard is not required to finish onboarding, and skipping it does not leave your tenant in an unconfigured state.

Manage your Escalation Router at any time from Governance > Configuration > Escalation Router.

Security

The Escalation Router includes several security features:

• Single-use tokens - Web response links can only be used once
• Expiration - Escalations expire after a configurable period (default: 7 days)
• Optional authentication - Require recipients to log in before responding via web portal
• Email validation - Responses are validated against expected recipients. In shared-sender mode this check is always on and cannot be turned off, because the sending mailbox is a general-purpose address that other people already write to

Related Topics

• Setup Wizard - Step-by-step setup guide
• Managing Escalations - View and manage pending escalations
• Configuration Options - Customize escalation settings
• Agents - Create and configure AI agents