Prove what was said on every call, not just the ones you sampled.
Regulated conversations need more than a happy caller: a disclosure read in full, a forbidden promise never made, an identity confirmed. VoiceLens checks each call against the phrases your rubric requires and forbids, and grades how reliable the transcript behind that verdict is.
Free to start · Works with any call platform · Same rubric for bots and people
How VoiceLens helps
From transcript to a scorecard your team acts on.
Required and forbidden phrases per rubric
Define the phrases that must be said and the ones that never may be, per template and per case branch, and VoiceLens verifies them on every call.
Transcript reliability is part of the grade
Low transcription confidence and speaker uncertainty are flagged as their own faults, so a verdict built on shaky audio is never presented as certain.
Evidence you can hand over
Every fault carries a timestamp and the transcript turn it came from; scorecards can be shared per call and analyses pushed to your systems via signed webhooks.
Fault detection
What VoiceLens catches on these calls
Every fault is timestamped and pinned to the transcript turn it came from, for AI and human agents alike.
Required phrase missed
A disclosure or confirmation the rubric requires was never said.
Forbidden phrase said
A phrase your rubric bans was used on the call.
Missed correction
The caller corrects a detail and the agent keeps the old one.
Low transcription confidence
Parts of the transcript are unreliable and should be trusted less.
Speaker uncertainty
Who said what is unclear, so the grade is flagged rather than assumed.
- Rubric templatesDefine what a good call looks like: slots, questions, required and forbidden phrases, case branching, thresholds, detector toggles, scoring weights, versions and AI drafts.
- Golden set and calibrationMark reference calls with the score you expect, re-score them against a rubric, and read the calibration report and fault-precision statistics.
- WebhooksReceive completed analyses, failures, threshold breaches and alerts as signed HTTP deliveries: event types, payloads, headers, HMAC verification in Node, .NET and Python, retries and troubleshooting.
Related solutions
Outbound AI voice agents
Detect dead air and repetition, score slot capture per attempt and compare agent versions on the same rubric.
Learn moreInbound support automation
Catch missed pivots and empathy gaps, track sentiment turn by turn and escalate flagged calls to your QA queue.
Learn moreHuman agent teams
Grade every call, coach with timestamped evidence and benchmark human agents against your bots on one rubric.
Learn moreSee what your QA process has been missing.
Create a workspace, drop in a recording or a transcript, and read the first scorecard in minutes.