Quality and Call QA: Calls
Objective. Open the Calls list, open one interaction, and read a single evaluation as one reviewer's scored read against a rubric - backed by evidence - not as objective truth.
The Calls screen is where interactions are listed and evaluated against the rubric. It answers 'what happened on this specific interaction and how was it scored?' - the atomic unit the whole loop is built on.
See it in the demo
What this screen answers
This is the interaction level: the list of calls (or cases) and the scored evaluation of any one of them. It answers 'on this specific interaction, what happened and how was it scored against our definition of good?' Sales uses one well-chosen call to make the loop concrete; a CSM uses it to inspect the evidence behind a score before trusting any aggregate built from many such scores.
The task
Filter the list to the team, agent or period in question, open one interaction, and read the score together with its evidence - the transcript or recording and the rubric items that drove it. Do not stop at the number. The task is to connect the score to the evidence and the rubric line that produced it, so you can judge whether the score is defensible, not just whether it is high or low.
How to read it
The classic misread is taking one call's score as an objective verdict on the agent. A single evaluation is one reviewer's read against the rubric on one interaction - a sample point, not a performance rating. Its meaning depends on who scored it, against which rubric version, and whether that reviewer is calibrated. Always read the score next to its evidence; a score you cannot trace back to a transcript line and a rubric item is not yet defensible. One good or bad call is not a trend.
Honest limits
A single call does not prove an agent's quality, cannot be generalised to the population, and does not show whether the rubric itself fits the customer. Any AI-assisted score here is a first pass that needs human review. Confirm with product which recording and transcription sources are supported and entitled; the demo shows the evaluation mechanics on seeded interactions.
Interpretation check
One call's score is one reviewer's read against the rubric on one interaction - a sample point backed by evidence, never an objective verdict on the agent.
In the field
KnackGlobal (RCM): inspecting the evidence behind one interaction's score - documentation or compliance handling - is exactly how a CSM shows that a QA number is traceable, not a black box.
Practical task
Open one interaction on the Calls screen and write a four-line note: the score, the two rubric items that most drove it, the piece of evidence you would cite, and one reason the score could be disputed - then state why one call is not an agent verdict.
Self-check
An agent is marked down on one call. Is that a performance problem?
Not on its own. One evaluation is a single sample point; a performance read needs an aligned, calibrated sample of enough calls, not one score.
What must you read alongside the score on an interaction?
The evidence - transcript or recording - and the rubric items that drove the score. A score you cannot trace to evidence and a rubric line is not yet defensible.
Two reviewers scored similar calls differently. What does the Calls screen alone tell you about who is right?
Nothing conclusive - that is a calibration question. The Calls screen shows each scored read; agreement is measured on the calibration screen.
If a score here was AI-assisted, what is its status?
A first-pass read that needs human review before it carries consequences; AI accelerates the evaluation, calibration and a human reviewer make it defensible.
Verified vs demo. Screen is live in the demo at /#/calls; exact controls and figures are SME-validation-pending.