AI Quality Auditor
A rubric you write, applied to calls once they are over. You set the sections, the weights and the pass mark; the model reads the call against them and returns a verdict with the quotes it scored on, which is what a sentiment line never gives you.
What gets read
read- Review is on for this workspace
- Review is on for voicemails
- Running on its own, not on request
- Longer than the 30 second floor
Product figures from the platform’s own defaults - not customer averages
How it works.
You write the scorecard
Sections such as Compliance, Opening, Discovery and Closing, each holding checkpoints of four kinds: scored out of ten, checklist yes, no or NA, critical pass or fail, and open-ended for the model to answer in prose.
Sampling picks what gets read
Every call that matches, a number at random per agent per day, a share of the run, or every Nth call. Narrow it by campaign, queue, list, agent, disposition or length, on a cadence with a daily AI spend cap if you want one.
Scoring, then the evidence for it
Checkpoints score, sections roll up to a weighted total against your pass mark, and a single critical failure sinks the review whatever that total came to. Coaching notes carry quotes, and a quote seeks the audio to the moment it came from.
Two speakers, kept apart
Agent and lead audio are split before any scoring runs, so a talkative lead cannot drag a score about agent behaviour. Talk-time share, interruption count and longest monologue fall out of the same pass.
One moment in every call.
Every call passes through the same seven. AI Quality Auditor is the lit one, and everything either side of it is a different page in this category.
- 01Number
the line it arrives on or leaves from
- 02Route
where the call is sent and to whom
- 03Dial
placing it, and how many at once
- 04Connect
the moment a person picks up
- 05Talk
while the conversation is happening
- 06Wrap
the code, the notes, the next step
- 07Record
what is kept, and where it is kept
The specifics.
8 facts- Checkpoints
- Scored 0 to 10, checklist yes/no/NA, critical pass or fail, open-ended text
- Weighting
- Sections carry their own weight, and the pass threshold is yours to set
- Critical items
- One failure ends the review at fail, whatever the weighted total reached
- Versioning
- Saving cuts a new version; past reviews stay tied to the rubric that scored them
- Sampling
- Four rules: all matching calls, N random per agent per day, a percentage of the run, or every Nth
- Trends
- Section averages week to week, and open-ended answers grouped into repeat gaps
- Access
- A QA-reviewer role in the role tree, scoped by the sharing rules you already run
- Versus analysis
- Call analysis returns transcription, sentiment and a suggested code, no rubric
What runs alongside it.
Volume calling is several controls working at once, not one dial turned up. These are the others.
More in Outbound & Compliance
3 capabilitiesBroadcasting, voice AI, QA scoring, calling rules and number health.
Send one recorded message to a whole list at once, with machine detection, a press-one path to a live agent, and every outcome written back to the record.
Answer rate per number, spam-label and CNAM status from carrier databases, a warm-up ramp for new numbers, and the thresholds that decide when one is rested.
A call step inside the mail cadence, with the outcome a rep logs against it choosing the next branch. Email and SMS as steps on the very same canvas.
The rest of the platform.
Five more categories, all on the same record and the same bill. Each card names three of its capabilities, so you can tell from here whether it is worth opening.
Outbound & Compliance
See ai quality auditor on your own floor.
Thirty minutes, your numbers and your data. We will set ai quality auditor up live and you can decide from the thing itself rather than from this page.
14-day trial · no card · migration included