CypherMonkeyAI omni-channel lab

A lifelong-learner lab on AI omni-channel agents, including voice and the contact center. Free to read. Nothing sold.

Speech analytics and quality · L2 · 20 min

A QA score needs a rubric

Learn in five minutes

  • Wikipedia's speech-analytics page says the technology can point to where an agent may need coaching and can monitor service on calls.
  • A number with no written rubric is a mood. Write the behavior you are scoring.
  • That page defines precision as the share of search results that are relevant, and recall as the share of relevant items the search returned.
  • Those two measures answer different questions. A high number on one is not a high number on the other.
  • This lab does not publish a contact-center quality score as a fact.

Write the rubric in words before you write a point value. If you cannot say what a miss looks like, you do not have a score.

This section HTML

Visual

T10-R3 diagram

Diagram SVGDiagram PNGDiagram HTML

Flow SVGFlow PNGFlow HTML

One behavior, one check

  1. Behavior: the reply quoted the note or refused.
  2. Miss: the reply added a date the note does not contain.
  3. Score that one behavior. Leave every other behavior unscored.
  4. Do not call the result a vendor quality score.

This section HTML

Quiz

What does this room require before a point value?

The rubric is the behavior in words.

What does the speech-analytics page say precision measures?

Recall is the other measure. They are not the same.

What quality score does this lab publish?

The room refuses a borrowed score.

This section HTML

Ecosystem

Rubric
The written behavior a score is about. Without it, the number has no meaning in this lab.

Sisters: Automatic speech recognition, Sentiment analysis, Quality rubrics, Personal data.

The live voice pipeline answers the person. Analytics reads a recording later. Keep those clocks apart.

A later review can add a rubric you wrote. It cannot borrow a customer's tape.

Config

  • Write the goal in one sentence.
  • Name the input and the allowed output.
  • Save the trace next to the result.

Troubleshooting

  • The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
  • The loop never stops. Set a step cap and stop when the tool result repeats.

This section HTML

Sources HTML

Sources and go deeper

sources.json