AI voice agent call QA data
Calls between AI voice agents and real callers, scored by people who have done the job: collections, lending, insurance and Medicare enrolment, healthcare front desks, home services, property and car dealerships. Each call is rated turn by turn for accuracy, compliance, tone and hand-off, with the exact moment it went wrong.
Specification
- Industries
- Collections, lending, insurance and Medicare enrolment, healthcare front desks, home services, property management and car dealerships
- Calls
- AI voice agents with real callers playing everyday customer situations, in English and Spanish
- Reviewers
- People with work experience in the industry of the call, from the US, the UK and Mexico
- Scores
- Accuracy, required disclosures and scripts, tone, task completion and hand-off to a human, per turn and per call
- Annotation
- Timestamped issues, the correct answer for each error, its severity and a written reason
- Quality check
- Several reviewers score each call on their own; hidden reference calls catch careless scoring
Example of the data
What you see, and the file you get.
| Turn 4 · agent | wrong payoff date | accuracy ✗ |
|---|---|---|
| Turn 7 · agent | skipped a required disclosure | compliance ✗ |
| Whole call | tone 4 / 5 | hand-off: too late |
| audio_url | industry | turn | dimension | score_1_5 | issue | reviewer |
|---|---|---|---|---|---|---|
| …/audio/call_0001.wav | collections | 4 | accuracy | 2 | gave the payoff date as 15 June; the account shows 12 June | R017 |
| …/audio/call_0001.wav | collections | 7 | compliance | 1 | required debt-collection disclosure not said | R017 |
| …/audio/call_0001.wav | collections | all | tone | 4 | R017 | |
| …/audio/call_0002.wav | healthcare_front_desk | 3 | task_completion | 3 | booked a new-patient slot for an existing patient | R042 |
| …/audio/call_0002.wav | healthcare_front_desk | all | hand_off | 2 | caller asked for a person twice before the transfer | R042 |
| …/audio/call_0003.wav | auto_dealer | 5 | accuracy | 5 | R009 |
Sample rows. Also as JSON or Excel.
Contact us to get the dataset
Tell us what you need. We walk you through the data, its formats and how it is checked on a short call.
What it is used for
Choosing a voice agent
Compare agents or versions on the same calls, scored by people who know the work.
Compliance checks
Find the calls where a required disclosure was skipped or a promise was made that should not be.
Judge and reward models
Teach an automatic call scorer to agree with people from the industry.
Release testing
Score every new prompt or model on the same calls before it reaches customers.
Questions
Who scores the calls?
People who have worked in the call’s industry, such as collectors, loan officers, insurance agents and front-desk staff, so a compliance miss is caught by someone who knows the rule.
What exactly is scored?
Each turn and each whole call: whether the facts were right, the required disclosures were said, the tone fit, the task was done and the hand-off to a human came at the right time.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples