AI to human handoff
AI Escalation Management: Handing Conversations to Your Team
When the AI should not finish a conversation, the right person gets it with the whole story attached.
Last updated 2026-10-28
AI escalation is the moment an AI agent stops and passes a conversation to a person. Managing it well comes down to three things: the AI escalates the right conversations, they reach the right person quickly, and that person can pick up without asking the customer to start over. In Savena the AI escalates inside the same conversation, attaches a brief of what the customer wants and what it already did, files the conversation under one of your escalation topics with a priority, and routes it by your assignment rules. Conversations the AI escalates are never billed.
When the AI escalates
- The customer asks for a person.
- Your knowledge base and connected data do not answer the question.
- The request needs an action you have switched off, or goes past a limit you set, such as a discount above the auto-issue cap.
- A Shopify order has already shipped and the customer wants a new address.
- The customer wants a refund on a Shopify order. The AI never issues those; a person does.
- Identity verification fails repeatedly.
- The issue is sensitive or complex enough that it should not decide alone.
What your agent gets with the handoff
The same conversation
The handoff happens in the thread the customer is already in, on the channel they used. Your agent sees every message the customer and the AI exchanged.
A brief from the AI
Why it escalated, what it already did, and a checklist of the changes the customer asked for, so the agent knows what is left to do before reading the thread.
A topic and a priority
You define escalation topics, such as "Damaged item" or "Wholesale inquiry", each with a priority. The AI files every handoff under one, and the topic's priority is the minimum the conversation gets.
The customer's history
The contact profile keeps the customer's conversations from email, live chat, WhatsApp, Instagram and Messenger together, with their lifecycle stage and notes.
Who gets the conversation
- Assign escalations round-robin, to whoever has the fewest open conversations, or leave them in the unassigned queue for the team to pick up.
- Send escalations to a default team first.
- Assign only to agents who are online, and skip agents who already have a set number of open conversations.
- When nobody is online, the conversation waits in the unassigned queue and the team is notified. Pending escalations arrive as one digest email per agent, ordered by priority, and an escalation nobody has touched is flagged again.
- On the live chat widget, the AI asks an anonymous visitor for an email before it escalates, so your reply reaches them even after they close the tab.
Seeing what gets escalated
AI agent analytics count escalations by topic, so you can see how many were damaged items or wholesale inquiries this week, and the inbox filters by topic. A topic that keeps growing usually means a gap in the knowledge base or an action worth switching on.
What it does not do
- No phone or SMS: escalations stay on email, live chat, WhatsApp, Instagram or Messenger.
- No SLA timers or breach alerts.
- Refunds on Shopify orders always go to a person.
Frequently asked questions
What is AI escalation in customer service?
AI escalation is when an AI agent hands a conversation to a human agent because it cannot or should not finish it alone. A good escalation keeps the conversation history and tells the human what the customer needs, so the customer does not repeat themselves.
How does Savena's AI decide when to escalate?
It escalates when the customer asks for a person, when your knowledge base and connected data do not cover the question, when the request needs an action you have switched off or exceeds a limit you set, when identity verification keeps failing, and for refunds on Shopify orders, which always go to a person.
Does the customer have to repeat themselves after a handoff?
No. The agent picks up in the same conversation with every message visible, plus a brief from the AI that says why it escalated, what it already did, and which changes the customer asked for.
Can a WhatsApp chat be escalated to email or phone and keep the thread?
In Savena the handoff stays in the WhatsApp conversation, so your agent replies on WhatsApp with the full thread in front of them. Savena has no phone or SMS channel. If the same customer also emails you, both conversations sit on their contact profile.
Can escalations go to a specific team?
Yes. You can send escalations to a default team first, then assign them round-robin or to the agent with the fewest open conversations, optionally only to agents who are online.
What happens if nobody is online when the AI escalates?
The conversation goes to the unassigned queue and your team is notified. Pending escalations are grouped into one digest email per agent, ordered by priority, and an escalation that sits untouched is flagged again.
Do escalated conversations cost anything?
No. Savena charges $0.99 only for conversations the AI resolves on its own. A conversation it hands to your team is not billed.
Setting it up: AI escalation to human agents in the help center. Escalation on messaging channels: WhatsApp shared inbox, Instagram support.
Let the AI finish what it can and hand over the rest
21-day free trial, no credit card required.
Start free trial