Extractors

Extractors turn a conversation into structured data. Define the fields you care about — name, phone, email, reason for calling, budget, preferred time, anything — and after each conversation the AI reads the transcript and fills them in. You get clean lead data, not just a transcript to read.

Where: open an agent → Extractors.

How it works

1. You define a list of fields — each one is a question about the conversation, with a type.
2. The conversation happens as normal. Extractors read, they don’t ask — they never change what the agent says.
3. When the conversation ends, the AI analyses the transcript and answers each question. A field that wasn’t covered in the conversation is left empty — nothing is guessed.
4. The answers are saved with the record and shown in Call history and the agent’s Captured Data tab.

This works on every channel — phone calls (inbound and outbound), the website chat widget, and WhatsApp.

Defining a field

Each extractor has:

• Field name — a short snake_case key, e.g. caller_name (typed spaces become underscores automatically). It’s the label on the captured value and the column header in Captured Data.
• Question / instruction — what to find out, e.g. “What is the caller’s full name?” or “Why did they call?”. The AI uses this to pull the answer from the transcript.
• Field type — one of three:
  • Open Question — a free-text answer from the transcript.
  • Yes / No — returns true or false.
  • Single Choice — the AI picks one of your options.
• Options — Single Choice only: at least two comma-separated values, e.g. Morning, Afternoon, Evening.

Each field also has an on/off toggle — switch one off to pause it without deleting it; disabled fields are simply skipped. Fields are listed (and shown as columns) in the order you add them.

Good fields to capture

• Lead basics: full name, phone, email — as Open Questions.
• Intent: reason for calling, service interested in — Open Questions; urgency works well as a Single Choice.
• Qualifying: budget, timeframe, location — whatever your sales/ops team needs.
• Outcome: a Single Choice like booked, needs callback, not interested, or a Yes / No like “did the caller agree to a follow-up?” — easy to scan and filter later.

Where the data goes

• Per conversation: an Extracted Data card on the call in Call history, alongside the transcript and (if enabled) the recording.
• Per agent: the Captured Data tab — one row per conversation, one column per field, so you can scan every lead in a single table.
• Into your systems: after-call Actions (send email, webhook, after-call SMS, WhatsApp message) can include captured fields — reference one as {{extracted.field_name}}, or the whole set as {{extracted_data}} in a webhook body — to push the lead straight into your CRM.

Best practices

• Keep the list focused. 4–6 well-chosen fields give you a table you’ll actually use; 15 vague ones mostly come back empty.
• Ask one thing per field. “What is the caller’s name?” beats “get the name, number and email” — split those into separate fields.
• Prefer Yes / No and Single Choice where you can — fixed answers are far easier to filter and report on than free text.
• Check a few calls in history to confirm the fields come back the way you expected, and tweak the questions if not.

Next steps