Skip to main content

Entity Extraction

Skip the manual data entry by automatically writing extracted data to your CRM

Written by Ethan Peterson-New

What is Entity Extraction?

Entity Extraction pulls specific pieces of information out of your recorded conversations and writes them into your CRM — with no post-call data entry from your reps.

You tell Siro what to look for: the competitor they mentioned, the price you quoted, their move-in date, the objection they raised. Siro finds it in every relevant conversation and sends it to the CRM field, task, or note you choose.

Who can set this up

Entity Extraction is admin-only. Find it under Admin → Entity Extraction in the left sidebar of the Siro web app. Reps and coaches don't have access.

What teams typically extract

  • Objections and concerns (price, timing, logistics)

  • Decision maker

  • Timeline — project start, delivery window, move-in date

  • Follow-up action items

  • Competitors or current vendor

  • Discounts and promotional terms

  • Quoted price or estimate

  • Financing discussed

  • Brand or material preferences

  • Order quantity

Create an extraction

Go to Admin → Entity Extraction and click New Extraction. There are four sections.

1. Extraction

  • Name — what this extraction is called.

  • Description — optional.

  • Conversation type(s) — which conversations this runs on. Siro only runs the extraction on recordings of the types you pick here.

⚠️ Pick at least one conversation type. In order to write to your CRM, at least one conversation type must be configured in the extraction’s settings.

2. Prompt

  • Prompt — describe, in plain language, what you want pulled out of the conversation.

  • Expected output — choose Single for a natural-language answer, or Multiple to get a comma-separated list.

The Test run tab lives here too. Use it before you go live (see below).

3. Output

Define what comes back. Most extractions need only one output; if you add several they're grouped together.

  • Output name — for example, Pain point.

  • Type — String, Number, Boolean, or Datetime.

  • Example output — a sample value. This steers the format Siro returns.

  • Word count limit — keeps long answers in check.

  • Predetermined outputs — restrict the result to a fixed list of options. Useful when the value has to match a CRM picklist.

4. Mapping

Where the value goes. You need a connected CRM integration before you can add mappings.

  • Where — which integration to write to.

  • Type — Field, Task, Note, or Activity.

  • Link to CRM Object Type — Default (auto), Appointment / Event, Opportunity, Contact, Account, or Lead.

  • Due in (days) — for tasks.

Then add a row for each value you're sending: Mapping from (one of your outputs) → Mapping to (the CRM field), and Content, which decides what happens to anything already in that field — Replace, Append, Prepend, or Don't Override.

The Content setting allows you to protect existing CRM data. Choose Don't Override for a field your team fills in by hand and you only want Siro to populate when it's empty; choose Append for a running log like a notes field. Append and Prepend add a dated header above each block they write, so a running log shows clear boundaries between entries.

Watch the field's length. If your CRM field has a character limit, Siro keeps the newest entries and drops the oldest ones to fit, so a long-running log will eventually lose its earliest entries. Use a long-text field for logs you want to keep in full.

If Content is left blank, it will behave as Replace.

One mapping per CRM field. Within a mapping, each CRM field can only be used once. Fields already used by another row are greyed out.

Append and Prepend only apply to free-text fields. On number, date or picklist fields Siro replaces the value instead.

Test before you go live

In the Prompt section, open the Test run tab and click Run. Siro runs your prompt against recent recordings and shows what it would have extracted from each one.

This is the fastest way to tighten a prompt before it starts writing to your CRM. We recommend a test run for every prompt change, so you always know what you’re getting.

What your reps see

On any recording, the CRM Autofill tab shows what was extracted. Each extraction appears as a card with a status badge — Extracting…, Extracted, Synced, or Extraction failed — above a table showing you the Field, Value and CRM mapping.

A banner at the top reports whether the recording is connected to a CRM record:

  • Recording linked to CRM as [name].

  • Recording not linked to CRM. Link it to sync extracted fields.

  • Maybe linked to [name] — accept to sync extracted fields, with an Accept match button.

Extracting and syncing are two different things. Siro extracts from the conversation either way, but it only writes to your CRM once the recording is linked to a CRM record. Unlinked recordings will create an extraction, but will not write it anywhere except within Siro.

When an extraction fails, the card shows a red Details: line with the error your CRM returned. That wording comes from the CRM rather than from Siro, so it can be technical, but it usually points at a field or permission problem on the CRM side. To retry, use Retry on the banner or Retry all extractions in a card's three-dots menu. Either one reprocesses the whole recording and runs each extraction again. On some older recordings, retrying isn't available — the Retry button doesn't appear and Retry all extractions is greyed out — because Siro can no longer reprocess those recordings.

Changing an extraction later

You can edit an extraction or add mappings at any time. Changes apply going forward:

  • Extraction runs once as each recording is processed, so an edit affects recordings processed from that point on.

  • Siro doesn't automatically re-run a changed extraction against previous recordings. Values already written to your CRM stay as they are.

    • Note: in certain instances, Siro Support or your CSM may be able to re-run historical extractions after an edit.

  • If you point a mapping at a different field, future values go to the new field. Recordings already processed keep their old values where they were written.

Important exception: If Siro reprocesses a recording — for example a backfill that regenerates summaries — extraction runs again on that recording and pushes to your CRM again. What happens to the value already in the field is decided by that mapping's Content setting. Use Don't Override to protect manually edited values.

Deleting an extraction removes it and all of its configuration permanently. Recordings that already ran it show Entity definition was deleted on the CRM Autofill tab.

Troubleshooting

  • The CRM Autofill tab is empty. It reads No entity definitions are configured for this conversation type when nothing is set up for that type.

    • Check the Conversation type(s) on your extraction.

  • Values show as Extracted but never Synced.

    • The recording isn't linked to a CRM record. Link it, or accept the suggested match on the banner.

  • An extraction failed. Read the Details: line on the card showing the error from your CRM.

    • This is usually a field type, required field, or permission problem. Fix it on the CRM side or within the extraction configuration, then retry. If retry isn't available on that recording, contact Support.

  • I can't add a mapping.

    • You'll see Connect a CRM integration before adding mappings. Set up the integration under Admin → Integrations first.

Have questions or need help?

Our Support team is here to assist. You can send us a message from the help menu in the mobile app, via the Get Help button at the bottom of the left sidebar on the web app, or send us an email at support@siro.ai.

Did this answer your question?