> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voiceinfra.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Summaries and Structured Data from Every Call

> Turn on post-call analysis to get a written summary of every conversation, and define a schema to pull the exact fields you care about out of the transcript.

Every call produces a transcript. **Analysis** turns that transcript into something you can act on: a written summary, and optionally a set of structured fields you define yourself.

Analysis settings are on the agent's **Analysis** tab. All fields are optional — leave them empty and the agent uses sensible defaults.

## Turning it on

**Enable Analysis** is the master switch. When it is off, the agent produces no summary and no extracted data.

<Note>
  Analysis is off by default on a new agent. If your call logs show no summary, this is the first thing to check.
</Note>

## The summary

Every analysed call gets a short written summary, generated from the transcript after the call ends.

**Custom Summary Prompt** replaces the default instruction — by default the agent produces a two to three sentence summary of the conversation. Use a custom prompt when you need something specific: a particular angle, a required level of detail, or a format your team expects.

The prompt is up to 5,000 characters.

<Tip>
  If your team reads summaries all day, write the prompt as an instruction to a colleague: *"Summarise for a support lead — lead with the customer's problem, then what was agreed, then any follow-up owed."*
</Tip>

## Structured extraction

This is how you get specific data out of a call rather than prose. It takes two settings working together.

<Steps>
  <Step title="Write the extraction prompt">
    **Structured Analysis Prompt** tells the agent what to look for. For example: *"From this conversation, extract the caller's name, the property address they asked about, whether they booked a viewing, and any budget they mentioned."*
  </Step>

  <Step title="Build the schema">
    **Structured Analysis Schema** defines the shape of the result using the visual builder. This is the list of fields you want back, and their types.
  </Step>

  <Step title="Enable analysis and test">
    With **Enable Analysis** on, run a test call, then open the call log and check the extracted data matches what you expect.
  </Step>
</Steps>

Both fields allow up to 5,000 characters.

<Warning>
  **The schema needs a prompt.** If you provide a schema but no extraction prompt, the agent has nothing telling it how to fill the fields. The editor warns you when this happens.
</Warning>

### Where the extracted data goes

Anything extracted is stored against the call and available from:

* The **call log**, alongside the transcript and summary
* **Webhooks**, in the payload sent when the call completes
* **Post-call emails**, if you have them configured
* **Functions that run after the call**, which receive the analysis as context

## What people extract

Extraction is generic — you define the fields, so it fits whatever you already track.

| Use case | Fields you might extract |
| - | - |
| **Lead qualification** | Company, budget, timeline, decision maker, fit score |
| **Appointment booking** | Name, requested date and time, service, confirmed or not |
| **Support triage** | Issue category, severity, affected product, resolved or escalated |
| **Orders** | Order number, product, quantity, delivery preference |
| **Sentiment and QA** | Caller sentiment, whether the issue was resolved, anything the agent handled poorly |
| **Compliance** | Whether required disclosures were read, consent given |

## Analysis works for chat too

These settings apply to chat sessions as well as voice calls. A widget or WhatsApp conversation produces the same summary and extracted fields, and they appear in the same call log.

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Why do my call logs have no summary?">
    Check that **Enable Analysis** is on. Analysis is off by default for a new agent.
  </Accordion>

  <Accordion title="Do I have to write a custom prompt?">
    No. With analysis on and no custom prompt, the agent writes a default two to three sentence summary. Custom prompts are for when the default is not the shape you need.
  </Accordion>

  <Accordion title="What is the difference between the summary and structured analysis?">
    The summary is written prose for a person to read. Structured analysis is a set of named fields for a system to consume — the same call, two different outputs.
  </Accordion>

  <Accordion title="Can I extract data mid-call instead?">
    Yes, and it is a different mechanism: use [conversation state updates](/docs/agents/system-tools) to capture values while the call is happening, or a workflow node's variables. Structured analysis runs after the call ends.
  </Accordion>

  <Accordion title="How much detail can I ask for?">
    Prompts allow up to 5,000 characters and the schema is yours to design. Very large schemas take longer and produce more chances for a field to come back empty, so start with the fields you will actually use.
  </Accordion>

  <Accordion title="Does analysis change what the caller experiences during the call?">
    No. Analysis runs after the conversation ends, so it does not affect the call itself or add latency.
  </Accordion>
</AccordionGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.