Your calls are your customers' words. We treat them that way.
You are about to hand a vendor recordings of your customers talking about their money, their health or their address. That deserves specifics rather than a badge, so here are the specifics — including which parts of this never leave our own infrastructure.
- 1. The data is yours, in the contract and in practice
- 2. It is never used to train a model
- 3. What never leaves our infrastructure
- 4. Isolated per company, encrypted throughout
- 5. Masked before it is stored, and not recoverable
- 6. Access that is actually enforced
- 7. An audit trail you can hand to a client
- 8. The infrastructure underneath
The data is yours, in the contract and in practice
The audio, the transcripts derived from it and every evaluation produced from those belong to your organisation. You can export all of it at any time, and when you leave you take it with you. We do not sell it, we do not share it, and we do not aggregate it into a benchmark product without asking you first.
It is never used to train a model
Your conversations do not train Google's models, ours, or anybody else's. Language requests go to Google's enterprise AI platform under terms that exclude customer data from model training. Nothing your customers said improves a model a competitor of yours will later use.
What never leaves our infrastructure
Two engines sit behind this product and it matters which does what. Transcription and scoring use Google's enterprise AI platform. Everything else — the acoustic analysis, the conversation metrics, the masking rules — runs on services we operate ourselves. That means the measurement of your audio is not a second vendor relationship for you to assess. Pitch, jitter, harmonics-to-noise, dead air, hold and ringback are computed by our own service, built on Praat and librosa, inside our own network. No third-party model sees the waveform for this, because nothing about measuring a waveform needs one.
Isolated per company, encrypted throughout
Every record is scoped to one company at the database level, and the scope is enforced on the server rather than in the interface. Data is encrypted in transit and at rest. A user's role decides what they can reach, and every route checks it independently of what the menu happens to show them.
Masked before it is stored, and not recoverable
The moment a transcript exists, a deterministic rule set removes card numbers, IBANs, CVV and PIN, e-mail addresses, phone numbers and long reference codes. It runs on our own servers, it is a pattern plus a real validity check — the Luhn algorithm for cards, the mod-97 check for IBANs — and it is reproducible: the same transcript masks the same way every time, which is what makes it a control rather than a filter. Masking happens before the transcript is stored, before it is scored and before it reaches a browser, and it cannot be undone. There is no encrypted copy of the untouched text, no support tool that opens it and no permission that reveals it. We do not hold what was masked. That is a deliberate trade and here is its cost: if a rule catches something it should not have, that text is gone for you as well as for us, and the conversation has to be re-uploaded. Report it and we fix the rule. The alternative — keeping every card number your customers read out, encrypted, against the day somebody needs one back — is not a trade we think you should have to accept. What it catches is structured data. It does not catch a name or a street address in ordinary prose, because nothing in the shape of that sentence marks it out. We say so plainly rather than let you believe the transcripts have been cleaned of everything.
Access that is actually enforced
Roles run from agent to platform administrator, and each one is a statement about which screens and which API routes are reachable. Two-factor authentication is available on every account, and sign-in with Google is supported where your organisation prefers it.
An audit trail you can hand to a client
Score changes, disputes, corrections, user changes and configuration changes are recorded with who did them and when. When an outsourcing client asks how a number was arrived at, the answer is a screen rather than a reconstruction.
The infrastructure underneath
Language processing runs on Google Gemini Enterprise, within Google Cloud's enterprise compliance programme — which includes SOC 2, ISO 27001 and GDPR commitments at the infrastructure layer. Those are Google Cloud's certifications covering the platform we build on; ask us directly about our own compliance position and we will tell you exactly where we stand rather than borrow theirs. The acoustic analysis service is ours and runs on our own infrastructure, which is why it is described separately above.
Send us your security questionnaire
We would rather answer forty specific questions than have you take a marketing page on trust. If procurement needs a document, ask and we will fill it in.
See it on your own calls
Send us a handful of recordings and your current scorecard. We will score them, show you the result next to what your team scored, and tell you honestly what you would and would not gain.
No automated demo. A real conversation, usually within one business day.