How we handle your customers' calls.

What happens to a recording

1. Arrival. A recording is uploaded from the browser straight to private object storage, or fetched from a link your phone system provides. Links must point to a public internet address; requests to private networks are refused.
2. Transcription. The audio is sent to OpenAI's speech-to-text API. Data sent through these paid APIs is not used to train the providers' models.
3. Redaction (if switched on). Before any language model reads the transcript, email addresses, card numbers, Aadhaar numbers, PANs, UPI IDs, IFSC codes, bank account numbers, OTPs, PINs and CVVs, and phone numbers are replaced with labels such as [PHONE]. The stored transcript never contains the originals; a recording kept for playback still does.
4. Analysis. The transcript is sent to OpenAI language models to separate speakers, summarise, score against your scorecards, re-check high-stakes answers and extract your fields.
5. Storage. The transcript, scores and extracted data are stored in your workspace. The recording is deleted as soon as processing finishes, unless your admin chose to keep recordings for playback.
6. Delivery. If you set up pipelines, alerts or the weekly summary, the results are sent to your CRM, Slack, email or webhook. Webhooks are signed so your systems can confirm they came from us.

Who can see what

• Access is by workspace role: owners and admins set things up; managers see their own team's calls; QA analysts review across teams; agents see their own calls; viewers have read-only access. These limits are enforced on every database query, not only hidden in the interface.
• To coach, agents are also shown short quoted lines from teammates' calls in the same company (how a teammate introduced themselves, or answered a price objection), with phone, card, account and ID numbers masked. They cannot open those calls.
• Admins can pause or remove anyone's access immediately; pausing signs the person out everywhere.
• Changes to members, roles, teams, agents, settings, integrations and scorecards, score reviews, coaching, disputes, exports and deletions are written to an audit log your admins can read. Anything our support team does in your workspace, such as changing your plan, is recorded there too.
• Our operator console shows companies, plans and processing health — not call content. We look at a specific call only when you ask us to for support.

Retention and deletion

• Recordings: deleted after processing by default, or kept for 7 days to a year for playback, as your admin chooses.
• Transcripts, scores and data: kept until you delete them, or deleted automatically after a period your admin sets (30 days to 5 years).
• Any single call can be deleted from its page. The workspace owner can delete the whole workspace from Settings: calls, recordings, scores, members and integrations are removed from the live system immediately. Database backups kept by our hosting provider expire on their normal schedule.

Technical safeguards

• All traffic uses HTTPS. Sessions are random, HTTP-only cookies; passwords are hashed with bcrypt; sign-in attempts are rate-limited.
• CRM tokens, webhook URLs and recording links that may contain credentials are encrypted at rest with AES-256-GCM.
• API keys are shown once and stored only as a SHA-256 fingerprint. Keys can be rotated or disabled instantly.
• Webhook destinations and recording links are checked against private, loopback and cloud metadata addresses before every request, and redirects are not followed blindly.
• Spreadsheet deliveries write values as plain text, so a phrase spoken on a call can never become a live formula in your sheet. CSV exports are protected the same way.
• Processing is idempotent: a retried request or a crashed worker never creates a duplicate call or a duplicate CRM record when your destination supports a call id field.

Where data is processed

The application and its database run in India; recordings are deleted after processing unless you keep them. Transcription and analysis are processed by OpenAI outside India; the transcript and results come back and are stored only in India.

India's Digital Personal Data Protection Act

When you use Convo Intelligence, you are the data fiduciary for your customers' personal data and we act as your data processor. The product gives you the controls the Act expects of you — purpose-limited processing, redaction, retention limits, deletion and an audit trail — but notice and consent for recording calls remain your responsibility. A data processing agreement is available for paid plans.

Reporting a security issue

If you believe you've found a vulnerability, please write to us from the contact page. We acknowledge reports within two working days and will keep you informed while we fix them.