The Analytics Tab configures what DialNexa extracts and analyzes automatically after each call completes. Post-call analysis runs as a background job using an LLM that reads the call transcript and populates structured fields on the call record. Use these fields to power dashboards, trigger workflows, filter Call History, and measure agent performance at scale.
How Post-Call Analysis Works
When a call ends:
- The full transcript is finalized.
- The post-call analysis job runs against the transcript using the extraction configuration defined in this tab.
- Extracted values are written to the call record.
- Values appear in Call History, the call detail view, and the API response for that call.
Analysis runs asynchronously and typically completes within 30 seconds of call end.
Extraction fields instruct the analysis LLM to identify and extract specific information from the transcript.
- Click Add Field in the Extraction Fields section.
- Configure the field:
| Property | Description |
|---|
| Field Name | The key used to store and retrieve this value. Alphanumeric and underscores only (e.g., appointment_date, reason_for_call). |
| Display Name | Human-readable label shown in the dashboard. |
| Type | The data type of the extracted value. See field types below. |
| Description | Instructions to the analysis LLM describing what to extract and how. Be specific. |
Field Types
| Type | Description | Example |
|---|
string | Free-text value extracted from the transcript | "Caller wants to reschedule to next Tuesday" |
number | Numeric value | 42, 1800.50 |
boolean | True or false based on whether something occurred | true if caller confirmed appointment, false if not |
enum | One of a predefined set of values | "new_customer", "existing_customer", "unknown" |
enum fields, define the allowed values in the Options list. The analysis LLM selects the closest matching option from the list. Include an unknown or other option to handle cases that do not fit defined categories.
Writing Good Field Descriptions
The description is the instruction given to the analysis LLM. Specific descriptions produce consistent results.
| Less Effective | More Effective |
|---|
"The reason for the call" | "The primary issue or request the caller described at the start of the call. Choose from: billing, technical support, account change, new inquiry, other." |
"Did it go well?" | "True if the caller explicitly confirmed they were satisfied or that their issue was resolved. False otherwise." |
"Appointment date" | "The date and time the caller agreed to for their appointment, as stated in the conversation. Return in ISO 8601 format (YYYY-MM-DDTHH:MM). Return null if no appointment was confirmed." |
Sentiment Analysis
Enable Sentiment Analysis
When enabled, post-call analysis includes a sentiment score for the overall call. The score is computed on a scale from -1.0 (very negative) to 1.0 (very positive), with 0.0 representing neutral.
Sentiment is assessed from the caller’s utterances in the transcript, not the agent’s responses.
Sentiment values appear in:
- The call detail view as a color-coded indicator
- Call History as a sortable/filterable column
- API responses under
analysis.sentiment
- A/B test comparisons if configured
Use sentiment as a quality signal rather than an absolute measure. Track average sentiment across agent versions and over time. A sudden drop in average sentiment after a prompt change is a signal to investigate.
Sentiment Granularity
Control whether a single overall sentiment score is computed or per-turn sentiment:
- Overall: One score for the entire call. Lower compute cost. Suitable for most use cases.
- Per-turn: A sentiment score for each caller turn. Useful for identifying the moment a call turns negative. Higher compute cost.
Custom Labels
Custom labels are categorical tags applied to calls based on transcript content. They are similar to extraction fields but are displayed as tags rather than structured data.
Add a Custom Label
- Click Add Label in the Custom Labels section.
- Enter:
| Property | Description |
|---|
| Label Name | The tag value applied to matching calls (e.g., escalation_requested, competitor_mentioned, payment_issue). |
| Description | Instructions to the analysis LLM for when to apply this label. |
| Label | Description |
|---|
escalation_requested | Apply if the caller asked to speak to a human agent or supervisor at any point. |
callback_requested | Apply if the caller asked to be called back or left a callback number. |
complaint | Apply if the caller expressed dissatisfaction, frustration, or made a formal complaint. |
Accessing Analysis Results
In the Dashboard
Open any completed call in Call History. The Analysis section shows all extracted field values, sentiment, and applied labels.
Via API
GET https://api.dialnexa.com/v1/calls/{call_id}
analysis object:
{
"call_id": "call_01hy...",
"analysis": {
"appointment_date": "2026-06-03T14:00:00",
"reason_for_call": "technical support",
"call_resolved": true,
"sentiment": 0.62,
"labels": ["callback_requested"]
}
}
Save Changes
Click Save to save to the current draft. New field configurations apply to calls completed after the draft is published. Existing call records are not retroactively analyzed with new fields.
Post-call analysis settings are versioned. If you publish a new version with different extraction fields, calls on the old version use the old field configuration and calls on the new version use the new configuration. Account for this when comparing analytics across versions.
