What Is Escalation Rate?

Domos Team - 4 min read - Updated May 2026

Escalation rate is the percentage of conversations an AI hands off to your team instead of resolving on its own. If an AI handles 100 resident or prospect interactions and passes 20 to a person, its escalation rate is 20%. It is the mirror image of containment rate - the share the AI resolves itself - so as one goes up, the other goes down.

It is one of the most revealing signals of whether an AI is built well.

Escalation rate in one line

It is the share of conversations the AI sends to a human - and the number that tells you whether it knows its own limits.

Why this matters for operators

Escalation rate is where the "how much can this really handle?" question gets answered. Industry benchmarks for AI customer support put best-in-class containment at 70-80% (with average deployments at 40-55%), according to Decagon - meaning the strongest systems resolve most interactions and escalate the rest. Domos sits in that top band: it handles 70-90% of interactions autonomously and escalates roughly 20% to a human.

But the most important point is counterintuitive: lower is not automatically better. Two numbers should make you nervous:

• A very high escalation rate (most conversations bounce to staff) means the AI is barely helping - you are paying for automation and still answering the questions yourself.

• A near-zero escalation rate is worse than it looks. An AI that never escalates is an AI that answers everything - including the sensitive cases and the questions it does not actually know, which is exactly when hallucinations reach your residents.

A tool that brags about handling "100% with no human needed" is not showing confidence. It is showing it has no off-ramp.

How a well-built system handles it

The right target is not the lowest possible escalation rate - it is the right escalations:

• Resolve the routine, escalate the rest. High containment on the repetitive volume, deliberate handoff on the cases that need judgment.

• Escalate on uncertainty, not just on keywords. When confidence is low, hand off instead of guessing.

• Route sensitive cases to people on purpose. Hardship, disputes, and fair-housing-sensitive questions are escalations by design, not failures.

• Escalate to the right person. Every use case should go to the right team or person - the property manager, or the relevant department.

A healthy escalation rate is a feature. It means the system knows the difference between what it should handle and what your team should.

Bottom line: the right handoffs, not the fewest

Judge an AI by whether it escalates the right conversations, not whether it escalates the fewest. A confident "I've got this" on every message is not a strength - knowing when to bring in a human is.

Frequently asked questions

What is a good escalation rate? There is no single number, but a well-built AI typically handles the large majority of interactions on its own and escalates a meaningful minority. Domos handles 70-90% autonomously and escalates roughly 20% - the cases that genuinely need a person.

Is a lower escalation rate always better? No. A near-zero escalation rate often means the AI is forcing answers it should not - including ones it made up - instead of handing off. The goal is the right escalations, not the fewest.

How is escalation rate related to containment rate? They are two sides of the same coin. Containment rate is the share an AI resolves on its own; escalation rate is the share it hands off. Every point of containment is a point less escalation.

Want to see how Domos decides what to handle and what to escalate?

Not ready? Start with the principle behind it