Quick answer
By Vozah Editorial·Last updated
AI Sales Call Scoring: How the Vozah Rubric Actually Works
AI sales call scoring uses a language model to read a call transcript and grade it against a written rubric. This page publishes the exact rubric Vozah uses: the nine dimensions, the weight of each one, the 0-100 scale and the arithmetic that turns nine scores into one number. It also shows what a transcript-only scorer cannot see, so you know how much weight to put on a score.
Quick answer: Vozah scores a call on nine dimensions, each from 0 to 100. Discovery carries 20 percent of the overall score; Value Proposition and Objection Handling carry 15 percent each; Opening Hook, Qualification, Talk-to-Listen Ratio and Closing carry 10 percent each; Pacing and Next-Step Clarity carry 5 percent each. The overall score is the weighted average, rounded to a whole number.
Fast-Scan Summary
| Dimension | What a high score looks like | Weight |
|---|---|---|
| Opening Hook | A specific, relevant reason for the call that earns the next 30 seconds; not a generic "how are you today" | 10% |
| Discovery | Open, layered questions that uncover the real problem and quantify it; digs past the first answer | 20% |
| Qualification | Surfaces decision process, stakeholders, timeline and budget context without interrogating | 10% |
| Value Proposition | Ties value to the pain the buyer described, in outcome terms rather than a feature list | 15% |
| Objection Handling | Acknowledge, clarify, reframe, advance; finds the real concern without getting defensive or caving | 15% |
| Talk-to-Listen Ratio | The rep listens more than they talk on a discovery-heavy call; no monologues | 10% |
| Pacing | Matches the buyer's energy, does not rush, controls the conversation without dominating it | 5% |
| Closing | Asks for a specific, low-friction next step with a concrete time | 10% |
| Next-Step Clarity | Both sides know what happens next, when, and who owns it | 5% |
The descriptions above are condensed from the rubric text the scoring model receives for each dimension. The weights add up to 100.
How a Call Gets Scored
There are two ways a call reaches the scorer, and they differ in one important detail.
- AI roleplay practice. The rep talks to a voice AI buyer in the Vozah simulator. The conversation is transcribed turn by turn, so the scorer receives a transcript with every line labeled REP or BUYER, plus the scenario title and buyer persona.
- Uploaded real call. On paid plans, a rep or manager uploads an audio file (MP3, WAV, M4A, MP4 or WebM, up to 4 MB). The file is transcribed as plain text without speaker labels, and the scorer is told to work out which speaker is the sales rep before scoring them.
- Dimension scoring. The model reads the transcript against the nine rubric descriptions and returns a 0-100 score and written feedback for every dimension, plus an overall coaching summary. The instructions ask it to be calibrated: most real reps land between 40 and 75, 85 and above is reserved for genuinely excellent moments, and below 40 marks clear misses. Feedback is meant to quote or reference what the rep actually said or failed to say.
- Composite roll-up. Vozah multiplies each dimension score by its weight, adds the results and divides by 100. The output is rounded to the nearest whole number.
- Validation. A scorecard is only saved if all nine dimensions come back with whole-number scores in range. An incomplete response is treated as a scoring failure rather than shown as a partial score.
If a practice call captures no speech from the rep, no model call is made and every dimension is recorded as 0 with a note to check the microphone.
A Worked Scorecard (Hypothetical Example)
The scores below are a hypothetical example written for this article to show the arithmetic. They are not from a real customer call.
Scenario: a rep cold calls a busy sales director. The rep opens with a specific reason, asks a few good questions but stops at the first answer, pitches features when the buyer says "we already have a tool," then books a vague follow-up.
| Dimension | Score (0-100) | Weight | Score x weight |
|---|---|---|---|
| Opening Hook | 72 | 10 | 720 |
| Discovery | 58 | 20 | 1,160 |
| Qualification | 61 | 10 | 610 |
| Value Proposition | 66 | 15 | 990 |
| Objection Handling | 49 | 15 | 735 |
| Talk-to-Listen Ratio | 55 | 10 | 550 |
| Pacing | 70 | 5 | 350 |
| Closing | 64 | 10 | 640 |
| Next-Step Clarity | 60 | 5 | 300 |
| Total | 100 | 6,055 |
Overall score: 6,055 divided by 100 is 60.55, which rounds to 61.
Two things stand out. First, Discovery moves the overall score twice as much as any 10-percent dimension: raising Discovery from 58 to 78 adds 4 points to the overall score, while the same 20-point gain on Pacing adds 1. Second, the composite hides the coaching priority. The single lowest score is Objection Handling at 49, and the feedback for that dimension is where the next practice session should start. Example feedback (also hypothetical): "When the buyer said they already use a tool, you listed three features. Ask what they would change about their current tool before positioning anything."
To try the same arithmetic with your own scores, download the editable sales rep scorecard. Its formulas use these exact weights and the same 0-100 scale.
Why These Weights
Discovery has the largest weight because most of the other dimensions depend on it: a value proposition can only be tied to the buyer's pain, and an objection can only be reframed, if the rep found out what the pain and concern actually are. Pacing and Next-Step Clarity carry 5 percent each because they overlap with other dimensions. Pacing partly overlaps with talk ratio, and next-step clarity overlaps with closing.
These weights are a product design choice, not a research finding. Vozah uses one set of weights for every call type today; it does not reweight cold calls, discovery calls and demos differently. If your team cares more about a different dimension, read the dimension scores directly instead of relying on the composite.
What Transcript-Based Scoring Cannot See
The scorer reads text. That sets clear limits on what any score means.
- Tone of voice. No audio features, such as pitch, volume or hesitation, reach the model. It cannot tell whether "sure, send it over" was warm or sarcastic unless the words themselves make it clear.
- Exact talk time. The Talk-to-Listen Ratio dimension is judged from how much the rep says in the transcript, not measured in seconds. The rubric's target is rep talk around 40-50 percent or lower on discovery-heavy calls. A talk-time study measures something different, so do not compare the two numbers directly.
- Speaking speed and pauses. Pacing is inferred from the flow of the conversation, such as rushing past answers or stacking questions. It is not a words-per-minute measurement.
- Speaker identity on uploads. Uploaded recordings are transcribed without speaker labels. If the transcript makes it hard to tell rep from buyer, the scores will be less reliable than scores from a labeled practice call.
- Deal context. A single transcript does not show what happened on earlier calls, what the buyer said by email, or whether the next step actually happened.
- Industry nuance. A generic rubric can misjudge a value proposition that depends on technical or regulatory knowledge. Managers in specialized markets should review those dimensions closely.
How to Check the Scores for Your Team
Vozah has not published a study comparing its scores with trained human raters, so the most useful check is one you run yourself:
- Pick five to ten recorded calls or practice sessions.
- Have two managers score each call independently on the same nine dimensions and 0-100 scale, using the rubric descriptions above.
- Compare the two managers with each other first. If they disagree widely, tighten the rubric definitions before judging the AI.
- Compare the AI's dimension scores with the managers' average. Note which dimensions agree and which do not; the dimensions that depend on audio, such as pacing, are the likeliest to diverge.
- Use the dimensions that agree for weekly coaching and keep a human review step for the rest.
Our guide to measuring sales training effectiveness covers how to connect score changes to real-call behavior without overclaiming results.
Practice Scoring Versus Real-Call Scoring
Practice scoring and real-call scoring do different jobs. A practice call against an AI buyer can be repeated: a rep can run the same cold calling scenario several times in a row and watch one dimension move. A real call cannot be redone, but it shows whether the behavior survives a live buyer. Teams usually use practice to build a behavior, then check uploaded real calls to confirm it holds.
Conversation intelligence platforms such as Gong and Chorus are built around recording and analyzing real customer calls across a whole deal. Vozah is built around practice plus individual call review. For a feature-by-feature view, see Vozah vs. Gong and Vozah vs. Chorus, and check each vendor's current documentation before deciding.
Manual Scorecards and the Calculator
Two other Vozah resources score calls by hand, and neither produces the same result as the product scorer:
- The sales rep scorecard template is a manual scorecard for hiring, ramp and reviews. It uses the product's nine dimensions, weights and 0-100 scale, but a manager fills it in, so the scores will differ from an automated scorecard for the same call.
- The sales score calculator is a quick interactive estimate. It uses six dimensions on a 0–10 slider and changes weights by call type, so its result will not match a Vozah scorecard.
Companion Reading
- Vozah AI sales coach, where scorecards and feedback appear in the product
- Uploaded sales call analysis, where you can see supported formats and a sample scorecard
- Sales rep scorecard template, the editable manual version
- Ideal talk-to-listen ratio, what talk-time studies measure
- Sales coaching questions for managers, the human review on top of a score
Create a free account and score a practice call →