Glossary
Escalation
Definition
Escalation is transferring a customer conversation from automation (or a frontline agent) to a human with more authority or context.
For an AI agent, escalation is the safety valve: when confidence is low, the topic is sensitive (legal threats, chargebacks, serious complaints), or the requested action is risky, the right move is handing the conversation to a human - with everything the agent learned attached.
A good escalation is not a failure; a bad one is. Good means the human receives the classified intent, the relevant order, what the agent already checked, and a suggested reply. Bad means the customer repeats everything to a person starting from zero.
Why it matters
Escalation policy is where AI support earns or destroys trust. An agent that never escalates will eventually improvise on a chargeback thread; one that escalates well makes your human team faster instead of busier.
How Resolvo HQ approaches it
Resolvo HQ routes on confidence and sensitivity: sensitive intents always go to a human, destructive actions never auto-resolve below the bar, and every escalation carries full context plus a drafted reply. You can also run entire channels in co-pilot mode where a human approves every send.
Frequently asked
What should always be escalated, no matter how good the AI is?+
Legal threats, chargeback disputes, serious complaints, and anything involving safety. These are relationship and liability moments where a human owns the judgment - the AI’s job is to arrive with the file prepared.
Are escalations billable in per-resolution pricing?+
They should not be - an escalation means the AI did not finish the job. Resolvo HQ never bills escalations or clarifying questions.
See it on your own tickets
Replay real tickets in a sandbox and watch the agent resolve them before you pay.