What every call produces
Every call on Finn produces three artifacts: an audio recording, a speaker-attributed transcript, and a set of post-call analysis values. All three are attached to the same call record. You do not enable them per call. Recording is on by default and can be turned off per Finn under Call Settings, or for the whole workspace under Settings → Compliance.
| Artifact | Contents |
|---|---|
| Recording | Audio of the full call. |
| Transcript | Text version, attributed to Finn or Caller, timestamped. |
| Metadata | Timestamps, caller ID, and similar call-level fields. |
| Post-call analysis | Every structured field you defined for the Finn. |
Where the logs live
| Page | Scope |
|---|---|
| Call History | Every call in the workspace, searchable. |
| Deployment Analytics | All calls for one deployment, with per-call drill-in. |
| Finn Analytics | Aggregate performance of one Finn across its deployments. |
| Live Deployments | Real-time call counts and the active call table. |
The per-call view
Click any row in a call table to open the Data Extractor Sidebar. This is the screen you will spend the most time in when debugging, because audio, transcript, and analysis sit side by side.
The sidebar shows:
- Caller info: name, phone, and custom fields carried over from the audience.
- Audio player with waveform scrubbing. Clicking a transcript line jumps the audio to that moment.
- Transcript, speaker-attributed and timestamped.
- Post-call analysis fields with their extracted values. See post-call analysis.
- AI analysis: sentiment, summary, outcome category.
- Telemetry: latency, ASR confidence, voicemail detection result, hangup cause.
- Tags applied to the call, manual or rule-based.
Keyboard navigation moves between calls without closing the sidebar.
| Key | Action |
|---|---|
| ↑ | Previous call |
| ↓ | Next call |
| Esc | Close sidebar |
You can click a transcript line to correct it. The edit updates the analytics record. If you opt in, corrections also feed back into ASR improvement.
Freshness
Analytics aggregates are cached. When you are looking at a number and it disagrees with what you just saw happen on a live call, the cache is the usual reason. The dashboard shows a Cached badge when data comes from cache, and Refresh forces a recompute.
| Data | Cache TTL |
|---|---|
| Deployment list | 5 minutes |
| Per-deployment summary | 1 minute |
| Per-call records | Never cached |
Per-call records are always live, so the sidebar never lags. Aggregates can.
While a deployment is still running, success rate is computed against the calls completed so far and will move as more calls finish. Do not draw conclusions from the first few dozen calls.
Getting artifacts out
| Route | What you get | Where you do it |
|---|---|---|
| Per-deployment CSV | Every call row with all fields as columns. | Dashboard, from the deployment. |
| PDF report | Top-line metrics, charts, notable transcripts. | Dashboard, from the deployment. |
| Webhook stream | Completed call events pushed to your endpoint in real time. | Configured under webhooks. |
CSV and PDF export are dashboard actions. There is no export endpoint. For programmatic access to call records, see the Calls API.
The call.analyzed event carries the post-call analysis fields in its payload, which is the usual way to move results into a warehouse or CRM without polling. Event payloads are listed under webhook events.
Retention and what gets deleted
Recordings and transcripts share the same retention window.
| Setting | Value |
|---|---|
| Default retention | 90 days |
| Maximum configurable retention | 7 years |
| After the window | Permanently deleted, no recovery |
There is no recovery path after the retention window closes. If you need call audio for a dispute or an audit that may open months later, raise the retention setting before the calls happen, not after. See retention.
PII redaction is opt-in under Settings → Compliance. It redacts SSNs, full credit card numbers, and similar values from transcripts and exported data. It does not touch the audio. If a caller reads a card number aloud, that audio still exists in the recording, and the only way to remove it is to delete or trim that recording yourself.
In two-party-consent jurisdictions you must disclose recording at the start of the call, either in the Finn's welcome message or through the built-in disclosure injection under Settings → Compliance → Recording Disclosure. See recording consent and compliance.
Related
- Post-call analysis for defining the structured fields shown in the sidebar.
- Monitoring for live deployment health.
- Troubleshooting for reading a bad campaign from its call logs.