Field Properties for AI-Powered Data Extraction
This documentation explains the recently added field properties that control how AI extracts and processes data from media content in Fields and Automations.
Overview
This article details four key field properties that enhance AI-powered data extraction from media content within Intercom's Fields and Automations. These properties—Prompt, Allowed Values, Other Values, and Not Applicable Values—work together to provide granular control over how the AI identifies, extracts, and standardizes information.
PROMPT
The Prompt field provides custom instructions to the AI assistant for extracting specific information from media content. It serves as the primary directive that guides the AI on what data to identify and extract from audio, video, or text content.
How It Works
When you create a field, you can define a prompt that describes what information should be extracted. This prompt is then used in automations when the field is selected for mapping. The prompt auto-populates from the field definition when you select a field in an automation, ensuring consistency across your workflows.
Important Notes
Changes to the prompt in a field definition will NOT automatically update existing automations that use that field
The prompt works in conjunction with other field properties (allowedValues, otherValues, notApplicableValues) to provide comprehensive extraction rules
Prompts should be clear, specific, and describe the exact type of information you want extracted
Use Cases
Extract specific entities: "Identify all company names mentioned in the conversation"
Classify content: "Determine the sentiment of the speaker (positive, negative, neutral)"
Extract structured data: "Find all dates and times mentioned in the transcript"
Identify topics: "List all product names or services discussed"
Best Practices
Be specific about what you want extracted rather than using vague descriptions
Include context about the format or type of data expected
Reference related field properties in your prompt when applicable
Test prompts with sample content to ensure they produce desired results
ALLOWED VALUES
Allowed Values defines a list of primary or preferred values that the AI should extract from content. This field helps standardize extracted data by constraining the AI to specific predefined options.
How It Works
Allowed Values is an array of strings that represents the expected or preferred values for a field. When configured, the AI will attempt to match content against these values. The behavior depends on two key settings:
1. Selection Mode (Single vs Multiple)
Single Mode: The AI must select only ONE value that best matches the content
Multiple Mode: The AI can select one or more values that apply to the content
2. Other Values Setting
When Other Values is disabled: The AI MUST ONLY use values from the allowedValues list (strict mode)
When Other Values is enabled: The allowedValues become "preferred values" and the AI MAY include other values if they're more accurate
Output Format
Single Mode: Returns a single value (e.g., "value1")
Multiple Mode: Returns comma-separated values (e.g., "value1, value2, value3")
Use Cases
Standardized categories: ["Sales", "Support", "Marketing", "Product"] for department classification
Status values: ["Active", "Pending", "Completed", "Cancelled"] for workflow status
Priority levels: ["Low", "Medium", "High", "Urgent"] for task prioritization
Product types: ["Software", "Hardware", "Service", "Consulting"] for product categorization
Best Practices
Use clear, distinct values that are unlikely to overlap in meaning
Order values by priority or frequency if applicable
Keep the list manageable (typically 5-15 values work best)
Use consistent naming conventions across similar fields
Consider how values will be used in filtering, reporting, or analytics
Relationship to Other Fields
Works with allowedValuesMode to control single vs multiple selection
Interacts with otherValues to determine strictness of value matching
Uses notApplicableValues as fallback when no matches are found
OTHER VALUES
The Other Values setting is a boolean flag that controls whether the AI can accept values outside the allowedValues list when extracting data from content.
How It Works
When Other Values is enabled (true):
The allowedValues list becomes a "preferred values" list
The AI can identify and include values not in the allowedValues list if they are more accurate or relevant
This provides flexibility for handling unexpected or alternative terminology
Example: If allowedValues is ["Sales", "Support"], the AI might also extract "Customer Success" if it appears in content
When Other Values is disabled (false):
The allowedValues list becomes strict and mandatory
The AI MUST ONLY respond with values from the allowedValues list
No values outside the list will be accepted
This ensures complete data standardization
Use Cases
Enable when you want to discover new categories or values that might appear in content
Enable when dealing with content that may use synonyms or alternative terminology
Disable when you need strict data standardization for compliance or reporting
Disable when working with predefined categories that must not be expanded
Best Practices
Enable Other Values during initial data collection to discover common values
Disable Other Values once you've identified all common values and added them to allowedValues
Consider your data quality requirements: strict standardization vs flexible discovery
Review extracted "other" values periodically to identify candidates for the allowedValues list
Relationship to Other Fields
Only meaningful when allowedValues is configured
Works in conjunction with allowedValuesMode (applies to both single and multiple modes)
When disabled, ensures strict adherence to allowedValues list
NOT APPLICABLE VALUES
The Not Applicable Values field specifies what the AI should return when no relevant values are found in the content being analyzed. This provides a consistent fallback response for cases where extraction cannot identify matching data.
How It Works
When the AI analyzes content and cannot find any values that match the allowedValues (or any relevant values if allowedValues is not set), it will return the value specified in notApplicableValues.
Default Behavior
If notApplicableValues is not specified or empty, the system defaults to "N/A"
The AI is instructed to return ONLY this value and nothing else when no match is found
This ensures consistent handling of missing or non-applicable data
Use Cases
Boolean fields: Use "false" or "No" when the condition is not met
Text fields: Use "Unknown", "Not Specified", or "N/A" when information is missing
Categorical fields: Use "Uncategorized" or "Other" when no category matches
Date fields: Use "Not Available" when dates cannot be extracted
Common Values by Field Type
Field Type | Common Values |
Boolean fields | "false", "No", "N/A" |
Text fields | "Unknown", "Not Specified", "N/A", "-" |
Categorical fields | "Uncategorized", "Other", "None" |
Status fields | "Not Applicable", "Pending", "Unknown" |
Best Practices
Choose a value that clearly indicates the absence of data
Use consistent notApplicableValues across similar fields in your system
Consider how these values will appear in reports, filters, and analytics
Use values that are distinct from your allowedValues to avoid confusion
Keep values concise and clear for end users
Relationship to Other Fields
Used as fallback when allowedValues matching fails
Works independently but complements the allowedValues system
Important for maintaining data completeness in extraction workflows
FIELD INTERACTIONS AND WORKFLOW
These four fields work in harmony to create a comprehensive data extraction system:
The Prompt provides the initial instruction on what to extract
Allowed Values defines the preferred or required values to look for
Other Values determines flexibility in accepting values outside the list
Not Applicable Values provides a fallback when nothing matches
Example Workflow
Consider a field for "Call Type" classification:
Prompt: "Identify the primary purpose or type of this call from the conversation"
Allowed Values: ["Sales Call", "Support Request", "Product Demo", "Follow-up"]
Allowed Values Mode: Multiple (to allow calls with multiple purposes)
Other Values: Enabled (to catch new call types like "Training" or "Onboarding")
Not Applicable Values: "Uncategorized"
In this scenario:
The AI will look for the four listed call types in the content
It can also identify and extract other call types if they appear
If multiple purposes are discussed, it will return a comma-separated list
If no call type can be determined, it will return "Uncategorized"
CONFIGURATION IN AUTOMATIONS
When creating or editing automations with Magic Prompt actions:
Field Selection: When you select a field that has these properties configured, they auto-populate in the automation
Override Capability: You can override field-level settings at the automation level for specific use cases
Field Mapping: The extracted values are automatically mapped to the selected field in your media library
Consistency: Using field-level prompts ensures consistency, but automation-level overrides allow flexibility
TROUBLESHOOTING
Common Issues and Solutions
Issue: AI returns values not in allowedValues list
Solution: Disable "Other Values" to enforce strict mode, or add the returned values to your allowedValues list
Issue: AI returns "N/A" too frequently
Solution: Review your prompt for clarity, check if allowedValues are too restrictive, or verify content actually contains relevant information
Issue: Multiple values returned when single mode is expected
Solution: Verify allowedValuesMode is set to "single" and that your prompt clearly indicates single value extraction
Issue: Prompt changes not reflected in existing automations
Solution: This is expected behavior - update automations manually or create new ones to use updated field prompts
SUMMARY
These four field properties provide powerful control over AI-powered data extraction:
Prompt: Guides what to extract
Allowed Values: Defines preferred/required values with single or multiple selection
Other Values: Controls flexibility in accepting values outside the list
Not Applicable Values: Provides consistent fallback for missing data
Together, they enable precise, standardized, and flexible data extraction from your media content while maintaining data quality and consistency across your organization.