Skip to main content
Data extractors analyze resolved conversations and extract structured data: customer sentiment, issue type, resolution status, and any other insight you define. They turn unstructured conversations into actionable data.

​
How it works

  1. You define an extraction schema (what data points to extract).
  2. You write optional instructions (how the AI should interpret conversations).
  3. You attach the extractor to a channel’s inbound settings.
  4. When a conversation resolves, the extractor reads the full transcript and outputs structured data matching your schema.

​
Creating a data extractor

  1. Go to Settings > Data Extractors.
  2. Click Create and choose a template or start blank.
  3. Configure:
FieldDescription
NameIdentifier for this extractor (e.g., “Customer Sentiment”, “Issue Resolution”).
DescriptionWhat this extractor does.
ModelThe AI model to use for extraction. Any of the available models can be used.
InstructionsOptional guidance for how the AI should analyze the conversation. For example: “Focus on the customer’s emotional tone and whether their issue was fully resolved.”
SchemaJSON Schema defining the output structure: the data points to extract.

​
Schema editor

The schema defines what the extractor outputs. Use the visual schema editor or write JSON Schema directly: 
{
  "type": "object",
  "properties": {
    "sentiment": {
      "type": "string",
      "enum": ["positive", "neutral", "negative"],
      "description": "Overall customer sentiment"
    },
    "satisfactionScore": {
      "type": "number",
      "description": "Customer satisfaction score from 1-10"
    },
    "issueType": {
      "type": "string",
      "description": "The primary issue category"
    },
    "resolved": {
      "type": "boolean",
      "description": "Whether the issue was resolved"
    },
    "keyIssues": {
      "type": "array",
      "items": { "type": "string" },
      "description": "List of key issues raised"
    }
  }
}

​
Templates

Lookfor provides templates to get you started:
TemplateExtracted fields
Customer SentimentSentiment, emotional tone, satisfaction score, key issues
Issue ResolutionIssue type, severity, resolved (boolean), resolution summary, follow-up required

​
Attaching to channels

Data extractors are configured in channel inbound settings:
  1. Go to Settings > Channels and open any channel (Storefront, Email, Instagram, or Facebook).
  2. In the Data Extraction section, select your extractor.
  3. Save.
The extractor runs automatically when conversations on that channel are resolved. You can also attach extractors to Campaigns in their inbound settings, so replies to campaign emails are extracted as well.

​
Viewing results

Extraction results are stored per conversation and can be accessed from the conversation detail in the Inbox.

​
Advanced settings

For reasoning models (GPT-5 family), you can configure:
SettingDescription
Reasoning effortHow much the model “thinks” before extracting (low, medium, high). Higher effort produces more accurate extractions but takes longer.
Reasoning summaryLevel of detail in the reasoning trace (concise, detailed).
VerbosityControls output length and detail (low, medium, high).