Is containment rate a good way to measure an AI receptionist?
No, not on its own. Containment measures the share of calls that never reached a human, which rewards a system for trapping people. A caller who gave up, hung up and called a competitor counts as contained. Measure resolution instead: the share of calls that ended with the outcome the caller wanted, plus the share of callers who had to contact you again within a couple of days about the same thing.
What containment actually counts
Containment rate is calls handled without transfer divided by total calls. Nothing in that formula asks whether anything useful happened. A caller who hung up in frustration at the second menu is contained. A caller who was told "someone will get back to you" and never heard from anyone is contained. A caller who booked the wrong service is contained.
The metric was popular because it maps directly to headcount savings in a call center. In a home services business, where a single missed booking costs far more than a few minutes of a CSR's time, optimizing it is close to backwards.
The two metrics that replace it
- Resolution rate. Calls that ended in a defined successful outcome divided by calls that had a resolvable intent. Successful outcomes are things you can verify in another system: a job on the schedule, a payment posted, a confirmed reschedule, a message delivered to a named owner.
- Repeat contact rate. Of calls the agent handled alone, the share where the same customer called, texted or submitted a form again within forty-eight hours about the same issue. This is the number that exposes fake containment.
Both require joining call data to operational records, which is exactly why this cannot be answered inside the phone system. The phone system knows the call ended; only the field service platform knows whether a job exists. That join is the point of revenue intelligence.
A denominator choice that changes everything
Do not divide by all calls. A meaningful share of inbound volume is spam, wrong numbers, vendors and existing customers calling about something the agent was never scoped to do. Including them makes every ratio look worse and hides real movement.
Define the denominator as calls with an intent inside the agent's scope, publish that definition, and report the excluded volume separately. Any metric whose denominator is not written down somewhere will eventually be argued about in a meeting nobody wins.
When containment is worth watching
It is a fine secondary indicator once resolution is stable. A sudden drop in containment with steady resolution usually means call mix changed. A sudden rise with falling resolution means something broke and the agent is now confidently handling calls it should be escalating. Read it as a change detector, never as a goal.
Topics: metrics · containment · resolution rate · measurement
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.