How Scoring Works

Your rubric says what to look for. This page says how the evaluator reads it — including the cases that surprise people the first time they see them, like a twelve-second call scoring 100%.

How a call score is calculated

Every criterion contributes its points to a total:

Call score = points awarded ÷ points available

A rubric worth 110 points where the agent earned 88 gives 80%. Nothing is weighted twice — a criterion’s weight is its Max Points.

One rule overrides the arithmetic. When a criterion marked Critical Error scores exactly 0, the whole call is set to 0% no matter what else the agent did. A critical criterion scoring 1 out of 10 does not trigger it — only a flat zero does. That is why critical flags belong on non-negotiables only, and why a rubric with many of them produces a wall of zeros.

What the evaluator applies on its own

You do not need to write any of the following into your criteria. The evaluator applies them to every rubric, in every language, whatever your process is. Adding them to your criterion text is harmless, but it earns you nothing.

Situation	What happens
The customer hung up, the line dropped, the recording cut	Requirements whose moment falls after the call ended are skipped and awarded in full
The situation a criterion describes never arose	The criterion is skipped and awarded in full
Two criteria cover opposite outcomes of the same moment	Only the one matching what happened is scored
The agent skipped something before the call ended	Still scored — an early ending never excuses it
The criterion judges how the call ended	Always scored
A criterion forbids a behavior	Full marks unless the agent can be quoted doing it
The customer describes something from an earlier call	Not evidence — only this recording counts

The rest of this page explains the ones that change how a number should be read.

When a criterion does not apply

A criterion applies only when the call gave the agent both the occasion and the opportunity to satisfy it.

Conditional criteria. A criterion that names its own situation — if the customer objects, when the customer declines — is skipped on calls where that never happened. An agent is never marked down for a step the call did not call for.
Mutually exclusive criteria. When two criteria cover opposite outcomes of the same moment — the customer accepted, the customer declined — only the matching one is scored.
Handling the call well does not make a criterion apply. If the agent resolved things so smoothly that the triggering situation never appeared, the criterion still does not apply.

A skipped criterion is awarded its full points, and its justification says so in the language of the call — which condition was absent, or where the call ended. That justification is how you tell a skipped criterion from a perfect one.

Calls that end early

Calls end before the agent reaches every step all the time. The customer hangs up, the line drops, the recording cuts mid-word, the call is transferred. When that happens, every requirement whose natural moment falls after the ending is skipped and awarded in full.

Two limits keep this from becoming a loophole.

Position, not outcome. Anything the agent had a turn to do before the call ended is scored normally. An ending at minute four never excuses something skipped at minute one. If there was a turn in which the agent could have complied, it counts.

The ending itself is still judged. How the agent treated the customer, how holds and silences were handled, who ended the call and in what manner, and every prohibition — all scored on the part of the call that exists.

Together they produce three quite different scores from three calls that all end in a hang-up:

What happened	Result
The customer hangs up during the greeting	100% — nothing the agent owed had come due yet
The customer hangs up after the agent completed the substance of the call	95–100% — only the closing steps were out of reach
The customer hangs up late, but the agent had already skipped a required step	59% — the omission happened while the call was still live, so it counts

The first row is the one that generates support tickets. It is not a bug: the agent did everything that had become due in the seconds available. If your process disagrees — if a call that short should not be scored at all — filter those calls out by duration rather than trying to make the rubric punish them.

Prohibitions are scored on evidence

Some criteria describe conduct that is forbidden rather than required: must not, never, it is prohibited to. These are scored the other way round from everything else.

Full points — nothing in the transcript shows the agent doing it
Zero — the agent can be quoted doing it

Absence of evidence is compliance, not a violation. A prohibition is never “not applicable”: not doing the forbidden thing is the behavior being measured.

Two things do not count as evidence, and both matter because prohibitions are usually flagged critical, so a mistake here costs the agent the entire call:

Hearsay. A customer saying “someone already called me about this yesterday” is an account of a different call. It cannot establish a violation in this one.
Facts this recording does not contain. How often the customer was contacted before, what another department promised, what was agreed on a previous call — if you cannot hear it here, it cannot be proven here.

If your process needs those rules enforced, they need data the call does not carry, and a call-scoring rubric is the wrong instrument for them.

How this appears in your reports

On a 30-call sample of ours, a criterion covering “the customer accepted” showed 97.8% in the scorecard. It genuinely applied on 12 of those calls. It was not the team’s strongest behavior — it was their least frequently tested one.

You do not have to work that out from the justifications. A skipped criterion is marked wherever it is shown:

Where	What you see
Expanded call detail	An N/A badge on the criterion, its points shown in grey rather than green
Expanded call detail, above the table	”3 of 18 criteria did not apply to this call”
Table tab, Score column	A small N/A count beside the percentage
Dashboard, Score by Criterion	applied to 12 of 30 calls · 18 N/A under the bar
Call Center Report scorecard	An Applied to column, also in the PDF export
Excel export	An Applicability column; Criterion % is left blank on skipped rows

The points are still shown on a skipped criterion, because they still count toward the call total. Only the colour changes, so the arithmetic continues to add up.

Two practical consequences:

Compare a conditional criterion against itself over time, not against your unconditional criteria. Its trend is meaningful; its rank in the list is not.
Read the Applied to column before treating a high result as a strength. A criterion at 97.8% that applied to 12 of 30 calls is telling you about call mix, not performance.

Calls analyzed before this was introduced carry no applicability at all. Those show no badge and no coverage — a blank rather than a zero, because “we did not record it” is not the same as “it applied everywhere”. A period that spans the change shows the count of skipped calls without the applied to X of Y fraction.

When a recording cannot be evaluated at all

This one is the opposite case, and easy to confuse with the one above.

If a transcript is empty, corrupted, or otherwise impossible to evaluate, every criterion scores 0 — which trips any critical criterion and produces a 0% call. So a 0% is worth one look before you read it as agent behavior:

What you see	What to check
0% with justifications describing real conduct	A genuine critical error
0% with justifications saying the transcription could not be evaluated	A failed or empty recording — check the Files tab
100% on a very short call	Criteria that did not apply, not a perfect call

Reading the distinction from the API

Each criterion in a call’s COPC detail carries an applicability field, evaluated or not_applicable, from the call detail endpoint:

{
  "criterion_id": "closing_summary",
  "justification": "The customer hung up before the agent could summarize the agreement.",
  "quote": "[04:12] Customer: I have to go — goodbye.",
  "applicability": "not_applicable",
  "score": 15,
  "max_points": 15,
  "is_critical_error": false
}

If you build your own dashboards, filtering out not_applicable rows before averaging gives you the per-criterion number the caution above warns about.

The report endpoint carries the same distinction in aggregate: every scorecard row reports evaluated_calls and not_applicable_calls alongside calls.

Next steps

Evaluation Rubrics — writing criterion descriptions that score the way you intend
Understanding Results — reading the Dashboard and Table views
Tips for Best Results — rubric design and interpreting scores