Quality reviews
The PBX can score every call with AI (see Call quality in AI agents and the AI console). Quality reviews adds the human side: supervisors score calls by hand with the same scorecards, compare their scores with the AI's, and calibrate so that every reviewer scores the same way. Agents read the reviews of their own calls and confirm them.
Open Quality reviews (/admin/quality-reviews). Reading needs reports:read (a supervisor whose role is limited to departments sees the agents of those departments), evaluating and calibrating need quality:write.
Evaluate a call
- Copy the call's ID from the call history and paste it into Call ID. When the call has an AI score, the AI's score of each item is shown under the fields.
- Pick the Scorecard whose weights apply, or keep The company's active scorecard.
- Score the items that apply from 0 to 100: greeting, script, sentiment, compliance and promises. Leave the others empty.
- Write a Comment for the agent and choose Save evaluation.
The total is weighted by the scorecard exactly like the AI score, and the AI's total at that moment is kept next to it. Each reviewer evaluates a call once; to change your scores later choose Edit in the list. Only the reviewer who made an evaluation can change or delete it.
The evaluations and the comparison with AI
Evaluations lists the reviews of the last 7, 30 or 90 days with the agent, the reviewer, the total, the AI total and whether the agent has read it.
Manual and AI scores shows, per agent, the number of evaluations and their average, the number of AI-scored calls and their average, and the Gap between the two (positive when people score higher than the AI). By week shows both averages week by week, so you can see an agent improve. A large gap usually means the scorecard's phrases or weights do not describe what your reviewers look for, or that the reviewers need calibrating.
Calibration
Calibration checks that reviewers agree.
- Under Calibration, enter a call's ID and a name and choose Start session.
- Each reviewer opens the session, chooses Score this call and saves an evaluation. The session is preselected in the form.
- Until you have scored the call yourself you only see how many reviewers did, so nobody copies the first score.
- Once you have scored it, the session shows every reviewer's total and their distance from the group's mean, and per item the mean, the lowest and highest score, the spread and the AI's score. The reviewers agree when every item and the total are within 10 points.
- Close session when you are done. A closed session can no longer be changed and is visible to every reader.
Calibration scores stay in their session: they do not appear in the agent's reviews or in the manual average.
For agents: My quality reviews
Every signed-in user has Quality reviews under Team (/my-quality). It shows the reviews of the user's own calls, newest first, with the score of each item and the reviewer's comment. I have read it confirms a review; the supervisor's report counts the confirmed ones.
Every evaluation, change, deletion, confirmation and calibration session is recorded in the activity log.
Working with the lists
The review list works like the Users list: the ... menu at the end of one of your own reviews has Edit and Delete. Below 640 pixels wide the table turns into cards with no sideways scrolling. A list that cannot be loaded shows the reason and Try again.