Related Articles
The Built-In Escalation Tool
Quick Reference Keywords: Escalation · Escalate to team · Human attention · Escalated badge · Resolve escalation · Escalation alerts · Alert recipients
Overview
Every AI Text Agent has a built-in escalation tool it can use at its own discretion to alert your team that a conversation needs human attention. Unlike Conditional Tools, which you configure for specific scenarios, the escalation tool requires zero setup on the agent - it's on by default for all text agents (global and journey) and acts as a safety net for situations you haven't configured a specific tool for.
The key thing to understand: escalating is an alert, not a handoff. When the agent escalates, your team gets notified and the conversation gets flagged - but the agent keeps working. It continues gathering information, answering questions, and pursuing its goal via text unless the contact explicitly says they don't want to keep texting. An escalation means "a human should look at this," not "the AI is giving up."
When the Agent Escalates
The agent is instructed to escalate when:
- The contact is aggressive or hostile
- The contact raises something sensitive - legal threats, safety concerns, a formal complaint
- The contact explicitly asks for a human
- The situation falls outside the agent's instructions and it has no guidance for handling it
And critically, only when no relevant alert was already sent. The agent can see the team alerts and escalations that already happened in the conversation (including alerts sent by your conditional tools), so it won't re-flag a situation your team already knows about unless it materially changes or gets worse.
What the agent does after escalating: It keeps helping. If a contact asks for a human, the agent alerts the team AND offers to keep assisting by text in the meantime - many contacts will still answer questions or book an appointment while your team is looped in. The agent only winds the conversation down if the contact explicitly says they don't want to continue over text.
Who Gets Alerted
Configure at: Settings > Text AI > Escalation Alerts
Default (no configuration needed): All account admins receive escalation alerts via in-app notification and email.
What you can customize:
- Alert Recipients - Choose which admins and staff members receive alerts. Every selected recipient always gets an in-app notification; per person, you can also enable:
- Email - Requires your account's email notifications to be configured
- SMS - Requires the team member to have an internal staff contact with a phone number (same requirement as the Internal SMS conditional tool action)
- Alert Assigned User (Contact Owner) - When enabled, the contact's owner also receives the alert, in addition to the recipients above
Note on the Agent role: Team members with the Agent role are not selectable as recipients. Agents can only see conversations for contacts they own, so an escalation alert about someone else's contact would point them to a conversation they can't open. If an Agent owns the contact, they can still be alerted through the Alert Assigned User option.
Every alert includes the reason - a specific explanation written by the AI of why it escalated (e.g. "Contact is disputing charges on their invoice and threatening to leave a negative review"). Alerts link directly to the conversation.
How Escalations Appear in the Interface
Conversations page:
- An "Escalated" badge appears next to the contact's name in the thread list and in the conversation header
- Hover over the badge to see the reason the AI escalated
- A "Resolve" button appears next to the badge in the conversation header
Resolving an escalation: An escalation is cleared in any of these ways:
- Click "Resolve" in the conversation header after you've reviewed the situation
- Reply to the contact manually - sending a manual message to an escalated contact automatically resolves the escalation (you've stepped in, which is the point)
- The contact's journey completes or ends - escalation state is tied to the current conversation
The agent is aware of resolutions. Once resolved, it knows a team member reviewed the situation, and it can escalate fresh if something new needs attention later.
In AI Text Logs:
Escalations show up as escalate_to_team tool executions in the agent's logs, including the AI's full reasoning. See Auditing AI Text Agents.
Escalation Behavior Details
One active escalation at a time: While an escalation is active (unresolved), the agent won't send duplicate alerts. If it tries to escalate again within 24 hours, no new alert goes out - the agent is told the team is already aware and to keep working.
Repeat escalations: If an escalation goes unresolved for more than 24 hours and the agent determines the situation still warrants attention, it can escalate again. The second alert is marked URGENT.
Follow-up protection: While a conversation is escalated, the agent's sequenced follow-ups are pushed out 48 hours. An uncomfortable conversation shouldn't get a chipper "just checking in!" three hours after the contact got upset. Custom follow-ups the agent explicitly scheduled with the contact (e.g. "I'll text you Friday") still send - those are commitments.
Turn Off AI After Second Escalation (optional, off by default): In Escalation Alerts settings, you can enable a two-strike behavior: if a conversation is escalated a second time without anyone resolving the first, the agent sends one final message letting the contact know a team member will reach out directly, then stops responding for that contact entirely (Text AI is turned off). Leave this off unless your team reliably acts on alerts - an imperfect AI response is usually better than silence.
Escalation Tool vs. Conditional Tools
This tool is NOT a substitute for properly configured agents. It's a safety net for the situations you haven't planned for yet.
Conditional Tools are strictly better for any scenario you can anticipate:
- They trigger reliably on the specific conditions you define, not at the AI's discretion
- They can execute any combination of actions - finalize journeys, turn off AI, add tags, send webhooks, route to other journeys, etc. - not just alert the team
- They give you precise control over recipients, message templates, and rate limits per scenario
The escalation tool can only do one thing: alert the team and flag the conversation. If you find it firing for the same type of situation repeatedly, that's a signal to build something better - see the next section.
After an Escalation: Close the Loop
When you get an escalation alert, there are two jobs - and most people only do the first one:
1. Take care of the contact. Open the conversation, read the reason and the recent messages, and step in however the situation calls for.
2. Fix the agent so this never needs to escalate again. Every escalation is the agent telling you: "I didn't have instructions for this." Once the dust settles, go edit the agent:
- Add instructions for the scenario. If the contact asked something the agent couldn't answer, add the answer (or the policy for handling it) to the agent's instructions or your knowledge base.
- Add a conditional tool with appropriate actions for the scenario. If a contact disputing an invoice should always alert your billing person AND get tagged AND pause the journey - build exactly that. See Conditional Tools.
Check the AI Text Logs for the escalation's {trigger.reason} and the surrounding conversation - it usually tells you exactly what trigger condition or instruction to write. Done consistently, this turns escalations into a to-do list for making your agents definitively better: each scenario gets handled properly and automatically the next time it comes up, and the escalation tool goes back to catching only the genuinely new stuff.
Disabling the Escalation Tool
You can turn the escalation tool off entirely with the master switch at Settings > Text AI > Escalation Alerts. This removes the tool from all of your text agents.
We don't recommend it. Without it, an agent facing a hostile contact or an uninstructed situation has no way to wave a flag - it will just keep doing its best, and nobody finds out until someone happens to read the conversation. If you're getting too many escalations, the better fixes are:
- Tighten your agents' instructions so fewer situations are "uninstructed"
- Add conditional tools for the recurring scenarios (which the agent sees, preventing duplicate escalations for the same situation)
- Trim the recipient list or channels instead of disabling the tool
Troubleshooting
Problem: Not receiving escalation alerts
Solution:
- Check Settings > Text AI > Escalation Alerts - verify the master switch is on and you're in the recipient list
- For email: verify your account's email notifications are configured, and check spam
- For SMS: verify the team member has an internal staff contact with a valid phone number
- Check the in-app notification bell - in-app alerts always send to selected recipients
Problem: Agent escalates too often
Solution:
- Review the escalation reasons in AI Text Logs - they tell you exactly what the agent felt unequipped for
- Add instructions or conditional tools for the recurring scenarios (see "Close the Loop" above)
- Remember the agent won't duplicate alerts for situations already flagged - frequent escalations almost always mean genuinely uncovered scenarios
Problem: Agent escalated but kept talking to the contact
Solution:
- This is by design - escalation is an alert, not a handoff. The agent continues helping unless the contact explicitly declines to continue by text
- If you want the AI to fully stop in specific scenarios, build a conditional tool with a Turn Off AI action for those conditions instead
Problem: Escalation badge won't go away
Solution:
- Click "Resolve" in the conversation header, or send the contact a manual message (either clears it)
- If it reappears, the agent escalated again for a new reason - check the tooltip and the AI Text Logs
Related Features
- Conditional Tools - Configure precise triggers and actions for specific scenarios
- AI Text Agents Overview - Main text agent guide
- Auditing AI Text Agents - Review escalations and agent reasoning in logs
- Global AI Text Agents 101 - How global agents serve CS and triage roles