Templates · September 3, 2026
Call quality scorecard: a rubric for reviewing AI and human sales calls
A ready-to-use scorecard for reviewing sales call transcripts, with criteria, a 1-5 scale, red flags and a weekly review routine for tuning an AI voice agent.

A call quality scorecard lets you review calls the same way every time, so you can tell whether a change to the script or prompt made things better or just different. The rubric below works for human reps and for AI voice agents. Copy it into a spreadsheet, score a sample of transcripts each week and track the average.
If you are tuning an agent, pair this with improving an AI voice agent from call transcripts.
The call quality scorecard
Score each criterion from 1 to 5.
| Score | Meaning |
|---|---|
| 1 | Failed or harmful |
| 2 | Weak, clear problems |
| 3 | Acceptable, room to improve |
| 4 | Good |
| 5 | Excellent, nothing to change |
Criteria
| # | Criterion | What to look for |
|---|---|---|
| 1 | Opening | Clear who, why and permission within 20 seconds; AI disclosure if you use it |
| 2 | Listening | Responds to what the person said, not to the script |
| 3 | Discovery | Asks relevant questions, one at a time, skips ones already answered |
| 4 | Accuracy | Facts are correct, nothing invented, says "I'll check" when unsure |
| 5 | Tone | Calm, plain, not pushy, matches the person's pace |
| 6 | Objections | Acknowledges, asks, answers, then accepts a no |
| 7 | Next step | Clear outcome: booked, transferred, follow-up set, or polite close |
| 8 | Flow | Natural pauses and turn-taking, no talking over the person |
| 9 | Compliance | Honours opt-outs, calling rules and disclosure |
| 10 | Wrap-up | Details read back, thanks, clean ending |
Maximum score is 50. Convert to a percentage if you prefer.
Red flags
Any of these should be fixed regardless of the total score.
- Invented a price, date, guarantee or fact.
- Ignored a request to stop or to be removed.
- Continued after a clear no.
- Denied being an AI when asked sincerely.
- Gave legal, medical or financial advice.
- Booked the wrong time, date or contact detail.
- Talked over the person repeatedly.
- Loop: repeated the same question or line more than twice.
AI-specific checks
| Check | Pass condition |
|---|---|
| Interruptions | Stops speaking and responds to the interruption |
| Silence | Checks in rather than filling it with long monologue |
| Wrong number or voicemail | Ends or leaves a message appropriately |
| Out-of-scope question | Says it will follow up and notes it |
| Names, emails and numbers | Reads back and confirms |
| Delay | Responses arrive quickly enough to feel natural |
Score sheet (copy and paste)
Call ID:
Date:
Reviewer:
Agent / rep:
Outcome (booked / transferred / follow-up / declined / no answer):
Opening 1 2 3 4 5
Listening 1 2 3 4 5
Discovery 1 2 3 4 5
Accuracy 1 2 3 4 5
Tone 1 2 3 4 5
Objections 1 2 3 4 5
Next step 1 2 3 4 5
Flow 1 2 3 4 5
Compliance 1 2 3 4 5
Wrap-up 1 2 3 4 5
Total (out of 50):
Red flags:
What went well:
One thing to change:
Weekly review routine
- Pull 10 to 20 calls: a few booked, a few declined, a few unanswered or short, and any that looked unusual in the summary.
- Read the transcript, not only the summary.
- Score each call using the sheet.
- Group the problems: opening, knowledge gaps, objections, flow.
- Choose the two biggest issues and change one thing at a time in the prompt or knowledge base.
- Test by calling your own phone, then compare the next week's average with this week's.
In Purple Voice, the Calls section keeps transcripts, AI summaries and recordings in one place, so this review does not require exporting anything.
Interpreting results
| Pattern | Likely cause | First fix |
|---|---|---|
| Low opening scores | Opening too long or vague | Rewrite the first line |
| Low accuracy scores | Missing or conflicting facts | Update the knowledge base |
| Low objection scores | No patterns in prompt | Add the top five from the objection cheat sheet |
| Good scores, few bookings | Offer or list problem | Review the list and the offer, not the agent |
| Good scores, poor next-step | Booking flow weak | Check calendar connection and read-back |
Tips for consistent scoring
- Write one sentence of evidence for any score of 1 or 2.
- Re-score two old calls each month to check you still agree with yourself.
- Do not let one great or terrible call set the tone for the week.
- Keep the scale the same for months, so trends mean something.
If you are still building the agent, the AI voice agent prompt template and the launch checklist help you start from a stronger place.
Try it
Run a small campaign in Purple Voice, open the transcripts in Calls and score them with this sheet. The Free plan includes one agent and 100 minutes a month, with no card. Start at voice.purplerocket.co.
Frequently asked questions
What is a call quality scorecard?
A short rubric used to rate each call on the same criteria, such as opening, listening, accuracy and outcome. It makes reviews consistent so you can see whether changes actually improve calls.
How many calls should I score each week?
Ten to twenty calls is a useful sample for most small teams, chosen across good, bad and ordinary outcomes. Start with all of them in the first week of a campaign.
Can the same scorecard be used for AI and human callers?
Yes. The criteria are about the conversation, not who held it. Add a few AI-specific checks such as handling interruptions and not inventing facts.
What score means a call is good?
That depends on your scale, but a simple rule is an average of 4 or higher with no red flags. Any red flag should trigger a fix regardless of the average.
Who should score the calls?
Someone who knows the offer and the goal of the call. For consistency, have a second person score a few of the same calls and compare.
- quality
- review
- transcripts