Document Extraction

Document Extraction

Overview

This skill provides a structured approach to extracting structured data from documents using ServiceNow Document Intelligence. It helps you:

• Configure and manage extraction templates in sn_doc_template for different document types (invoices, contracts, forms, purchase orders)
• Submit documents for extraction via the sn_doc_intelligence_extraction pipeline
• Define field mappings in sn_doc_intelligence_field_map to map extracted data to ServiceNow table fields
• Review and validate extraction results from sn_doc_intelligence_extraction_result
• Set up validation rules to ensure data quality and flag low-confidence extractions
• Handle extraction failures and retrain models for improved accuracy

When to use: When you need to process documents at scale (batch invoice processing, contract metadata extraction, form digitization), when manually entering data from documents is error-prone or time-consuming, or when setting up new document processing pipelines.

Plugin required: com.sn_doc_intelligence

Prerequisites

• Roles: sn_doc_intelligence_admin, sn_doc_intelligence_user, or admin
• Access: Read/write access to sn_doc_intelligence_extraction, sn_doc_template, sn_doc_intelligence_field_map, and sys_attachment tables
• Knowledge: Understanding of the document types to be processed, their field layouts, and the target ServiceNow tables for extracted data
• Plugin: Document Intelligence (com.sn_doc_intelligence) must be activated
• Documents: Source documents must be in supported formats (PDF, PNG, JPG, TIFF)

Procedure

Step 1: Review Available Extraction Templates

Check which document extraction templates are already configured in your instance.

Using MCP (Claude Code/Desktop):

Tool: SN-Query-Table
Parameters:
  table_name: sn_doc_template
  query: active=true
  fields: sys_id,name,description,document_type,target_table,state,extraction_model,confidence_threshold,field_count
  limit: 50
  order_by: name

Using REST API:

GET /api/now/table/sn_doc_template?sysparm_query=active=true^ORDERBYname&sysparm_fields=sys_id,name,description,document_type,target_table,state,extraction_model,confidence_threshold,field_count&sysparm_limit=50&sysparm_display_value=true

Common template types:

Step 2: Create or Configure an Extraction Template

If no template exists for your document type, create one.

Using MCP:

Tool: SN-Create-Record
Parameters:
  table_name: sn_doc_template
  fields:
    name: "Vendor Invoice Template"
    description: "Extracts header and line-item data from vendor invoices for accounts payable processing"
    document_type: invoice
    target_table: sn_proc_invoice
    confidence_threshold: 0.85
    active: true
    state: draft

Using REST API:

POST /api/now/table/sn_doc_template
Content-Type: application/json

{
  "name": "Vendor Invoice Template",
  "description": "Extracts header and line-item data from vendor invoices",
  "document_type": "invoice",
  "target_table": "sn_proc_invoice",
  "confidence_threshold": "0.85",
  "active": "true",
  "state": "draft"
}

Step 3: Define Field Mappings

Map document fields to ServiceNow table columns for each extraction template.

Using MCP:

Tool: SN-Create-Record
Parameters:
  table_name: sn_doc_intelligence_field_map
  fields:
    template: [template_sys_id]
    source_field: "Invoice Number"
    target_table: sn_proc_invoice
    target_field: number
    field_type: string
    required: true
    validation_regex: "^INV-[0-9]{4,10}$"
    order: 100

Repeat for each field mapping:

Using REST API:

POST /api/now/table/sn_doc_intelligence_field_map
Content-Type: application/json

{
  "template": "[template_sys_id]",
  "source_field": "Invoice Number",
  "target_table": "sn_proc_invoice",
  "target_field": "number",
  "field_type": "string",
  "required": "true",
  "validation_regex": "^INV-[0-9]{4,10}$",
  "order": "100"
}

Step 4: Submit Documents for Extraction

Create an extraction request to process a document through the pipeline.

Using MCP:

Tool: SN-Create-Record
Parameters:
  table_name: sn_doc_intelligence_extraction
  fields:
    template: [template_sys_id]
    source_document: [attachment_sys_id]
    document_type: invoice
    state: submitted
    priority: 3
    requested_by: [user_sys_id]
    short_description: "Extract data from vendor invoice INV-2026-0456"

Using REST API:

POST /api/now/table/sn_doc_intelligence_extraction
Content-Type: application/json

{
  "template": "[template_sys_id]",
  "source_document": "[attachment_sys_id]",
  "document_type": "invoice",
  "state": "submitted",
  "priority": "3",
  "requested_by": "[user_sys_id]",
  "short_description": "Extract data from vendor invoice INV-2026-0456"
}

For batch processing, submit multiple documents:

# Iterate over attachments and create extraction records
for attachment_id in [list_of_attachment_sys_ids]; do
  curl -X POST "https://[instance].service-now.com/api/now/table/sn_doc_intelligence_extraction" \
    -H "Content-Type: application/json" \
    -d "{\"template\":\"[template_sys_id]\",\"source_document\":\"$attachment_id\",\"document_type\":\"invoice\",\"state\":\"submitted\"}"
done

Step 5: Monitor Extraction Progress

Track the status of submitted extraction requests.

Using MCP:

Tool: SN-Query-Table
Parameters:
  table_name: sn_doc_intelligence_extraction
  query: state=submitted^ORstate=processing^ORstate=review
  fields: sys_id,number,short_description,state,template,document_type,source_document,confidence_score,sys_created_on,error_message
  limit: 50
  order_by: sys_created_on

Using REST API:

GET /api/now/table/sn_doc_intelligence_extraction?sysparm_query=state=submitted^ORstate=processing^ORstate=review^ORDERBYDESCsys_created_on&sysparm_fields=sys_id,number,short_description,state,template,document_type,source_document,confidence_score,sys_created_on,error_message&sysparm_limit=50&sysparm_display_value=true

Step 6: Review Extraction Results

Examine the extracted data and confidence scores for each field.

Using MCP:

Tool: SN-Query-Table
Parameters:
  table_name: sn_doc_intelligence_extraction_result
  query: extraction=[extraction_sys_id]
  fields: sys_id,field_name,extracted_value,confidence_score,validation_status,mapped_field,needs_review
  limit: 100
  order_by: order

Using REST API:

GET /api/now/table/sn_doc_intelligence_extraction_result?sysparm_query=extraction=[extraction_sys_id]^ORDERBYorder&sysparm_fields=sys_id,field_name,extracted_value,confidence_score,validation_status,mapped_field,needs_review&sysparm_limit=100&sysparm_display_value=true

Evaluate results against thresholds:

Step 7: Validate and Correct Extracted Data

For fields that need review, update the extraction results.

Using MCP:

Tool: SN-Update-Record
Parameters:
  table_name: sn_doc_intelligence_extraction_result
  sys_id: [result_sys_id]
  data:
    extracted_value: "INV-2026-0456"
    validation_status: validated
    needs_review: false
    reviewer_notes: "Corrected invoice number; OCR misread '0' as 'O'"

Using REST API:

PATCH /api/now/table/sn_doc_intelligence_extraction_result/[result_sys_id]
Content-Type: application/json

{
  "extracted_value": "INV-2026-0456",
  "validation_status": "validated",
  "needs_review": "false",
  "reviewer_notes": "Corrected invoice number; OCR misread '0' as 'O'"
}

Step 8: Complete Extraction and Populate Target Record

Once all fields are validated, mark the extraction as complete and create or update the target record.

Using MCP:

Tool: SN-Update-Record
Parameters:
  table_name: sn_doc_intelligence_extraction
  sys_id: [extraction_sys_id]
  data:
    state: completed
    work_notes: "Extraction validated. 12/14 fields auto-accepted (>95% confidence). 2 fields manually corrected. Target record created in sn_proc_invoice."

Create the target record with extracted data:

Tool: SN-Create-Record
Parameters:
  table_name: sn_proc_invoice
  fields:
    number: "INV-2026-0456"
    vendor: [vendor_sys_id]
    invoice_date: "2026-03-15"
    due_date: "2026-04-14"
    invoice_amount: "15750.00"
    po_number: "PO0012345"
    state: pending
    source: document_intelligence
    work_notes: "Record created via Document Intelligence extraction [extraction_number]"

Using REST API:

PATCH /api/now/table/sn_doc_intelligence_extraction/[extraction_sys_id]
Content-Type: application/json

{
  "state": "completed",
  "work_notes": "Extraction validated. Target record created."
}

Tool Usage

MCP Tools Reference

REST API Reference

Best Practices

• Template per document type: Create separate templates for each document format; a single template for all invoices will underperform compared to vendor-specific templates for high-volume vendors
• Set appropriate thresholds: Start with a confidence threshold of 0.85 and adjust based on error rates; lower thresholds increase throughput but require more manual review
• Training data: Provide at least 20-30 sample documents per template for optimal extraction accuracy
• Validation rules: Use regex patterns, reference lookups, and cross-field validation (e.g., total = subtotal + tax) to catch extraction errors automatically
• Batch processing: Submit documents in batches during off-peak hours to avoid performance impact on the instance
• Monitor accuracy: Track extraction accuracy metrics over time; retrain models when accuracy drops below 90%
• Handle exceptions: Create extraction tasks in sn_doc_intelligence_task for documents that fail extraction, routing them to data entry staff
• Secure handling: Ensure document attachments comply with data retention policies; delete temporary extraction artifacts after processing

Troubleshooting

"Extraction template not found"

Cause: Template is inactive or the document type does not match Solution: Query sn_doc_template with active=true to see available templates. Verify the document_type field matches your submission.

"Low confidence scores across all fields"

Cause: Poor document quality (low resolution, skewed scan, handwritten text) or template mismatch Solution: Check document resolution (minimum 300 DPI for OCR). Verify the document matches the template's expected layout. Consider creating a new template for non-standard formats.

"Field mapping validation fails"

Cause: Extracted value does not match the validation regex or target field type Solution: Review the validation rule in sn_doc_intelligence_field_map. Update the regex to accommodate legitimate variations (e.g., different invoice number formats across vendors).

"Extraction stuck in processing state"

Cause: Document Intelligence engine timeout or processing queue backup Solution: Check the sn_doc_intelligence_extraction record for error messages. Verify the Document Intelligence engine is running: navigate to Document Intelligence > Dashboard. Re-submit the extraction if needed.

"Duplicate extraction records"

Cause: Same document submitted multiple times Solution: Before submitting, query sn_doc_intelligence_extraction with source_document=[attachment_sys_id]^state!=failed to check for existing extractions.

Examples

Example 1: Invoice Data Extraction

Input: PDF invoice from Acme Corp uploaded to ServiceNow

Process:

1. Identify template: "Vendor Invoice Template" (sys_id: abc123)
2. Submit extraction with attachment reference
3. Results: 14 fields extracted, average confidence 0.92
4. 2 fields below threshold: PO Number (0.78), Line Description (0.81)
5. Manual review confirms PO Number correct, Line Description corrected
6. Target invoice record created in sn_proc_invoice

Example 2: Contract Metadata Extraction

Input: Signed MSA PDF for legal review

Process:

Tool: SN-Create-Record
Parameters:
  table_name: sn_doc_intelligence_extraction
  fields:
    template: [contract_template_sys_id]
    source_document: [attachment_sys_id]
    document_type: contract
    state: submitted
    short_description: "Extract metadata from Acme Corp MSA"

Fields extracted: Contract parties, effective date, term length, total value, governing law, auto-renewal flag, notice period. Populates ast_contract record automatically.

Example 3: Batch Processing Tax Forms

Input: 50 W-9 forms uploaded for vendor onboarding

Process:

1. Query all unprocessed W-9 attachments
2. Submit batch extraction using W-9 template
3. Monitor: 47 extracted successfully, 3 require manual review (handwritten entries)
4. Create vendor records in core_company with extracted TIN, legal name, address
5. Route 3 exceptions to data entry team via sn_doc_intelligence_task

Query for batch status:

Tool: SN-Query-Table
Parameters:
  table_name: sn_doc_intelligence_extraction
  query: template=[w9_template_sys_id]^sys_created_on>=2026-03-19
  fields: sys_id,number,state,confidence_score,error_message
  limit: 50

Related Skills

• document/smart-documents - Manage document templates, versioning, and automated generation
• legal/contract-analysis - Analyze extracted contract data for risks and terms
• legal/contract-obligation-extraction - Extract obligations from contract documents
• procurement/invoice-management - Process extracted invoice data through AP workflows
• development/data-import - Bulk import extracted data into ServiceNow tables