How AI judges score each call
Every call is scored against your written rules, one rule at a time.
Scoring one rule
A judge reads the whole call against one written rule, then gives pass or fail with the moment it turned.
What a rubric isRule: before booking, read the address back and wait for a yes.
AgentSo that’s 14 Mill Lane, AB12 3CD. Is that right?
CallerYes, that’s it.
The address was read back in full and the caller confirmed it before the booking.
What a mark is built on
Written so a person could make the same call.
- 1
Rules in plain words
Start from ours, or write your own. Each rule says what counts as a pass.
- 2
The line behind it
Every pass or fail quotes the transcript and gives the time in the call.
- 3
You can overturn it
Disagree with a mark and overturn it, with a note saying why.
Checking the judges
Judges can be wrong. We will publish how often they agree with people, and the method.
- People mark a sample. Reviewers will mark the same calls without seeing the judge’s answer.
- Agreement per rule. A rule the judge gets wrong too often will be flagged on your report.
- The method first. We will publish the method, sample size and figures before quoting any number.
Run a test on your own agent.

- test calls
- 506
- passed
- 96.8%
- median reply
- 0.94 s
Holds the new day when a caller changes their mind.