Document Types

The Email Ingestion service automatically classifies incoming documents into specific types for appropriate processing and NetSuite record creation.

Supported Types

Vendor Bill

NetSuite Record: vendorbill

A vendor bill is an invoice FROM a vendor/supplier TO your company. You are the buyer receiving this invoice and will need to pay it.

Key Indicators:

• "Invoice" or "Bill" in header
• Your company listed as "Bill To" or "Ship To"
• Supplier/vendor letterhead and branding
• Payment terms (Net 30, Due on Receipt, etc.)
• Items or services you purchased

Extracted Fields:

Line Item Fields:

interface VendorBillLineItem {
  itemName: string;
  description?: string;
  quantity: number;
  rate: number;
  amount: number;
  taxCode?: string;
}

Customer Invoice

NetSuite Record: invoice

A customer invoice is an invoice FROM your company TO a customer. You are the seller billing the customer.

Key Indicators:

• Your company logo and letterhead
• Customer name in "Bill To" section
• Your terms and conditions
• Your bank details for payment

Extracted Fields:

Purchase Order

NetSuite Record: purchaseorder

A purchase order is a request to purchase goods or services. This is NOT an invoice - it's created before the purchase is fulfilled.

Key Indicators:

• "Purchase Order" or "PO" prominently displayed
• PO number (e.g., "PO-2024-001")
• Delivery/shipping instructions
• No payment terms (yet)
• Status indicators (Draft, Pending, Approved)

Extracted Fields:

Expense Report

NetSuite Record: expensereport

An expense report is a collection of receipts and expenses submitted by an employee for reimbursement.

Key Indicators:

• Multiple receipts or transactions
• Employee name and department
• Expense categories (Travel, Meals, Office Supplies)
• Reimbursement request form
• Approval signatures

Extracted Fields:

Expense Item Fields:

interface ExpenseItem {
  date: string;
  merchant: string;
  category: string;    // 'travel' | 'meals' | 'supplies' | 'other'
  description?: string;
  amount: number;
  currency: string;
  receiptAttached: boolean;
}

Receipt

No NetSuite Record (may attach to expense report)

A receipt is a simple proof of purchase, typically from a retail transaction.

Key Indicators:

• Single transaction
• Point-of-sale format
• Store name and address
• Date, time, and transaction ID
• Payment method shown

Extracted Fields:

General Request

No NetSuite Record

This category is used when:

• The email has no attachments
• The document type cannot be determined
• The confidence score is too low
• The document doesn't match any known type

Handling:

• Sent to the review queue
• No automatic processing
• Manual classification required

Classification Process

AI Classification Pipeline

flowchart TD
    A[Email Received] --> B{Has Attachments?}
    B -->|No| C[General Request]
    B -->|Yes| D[Extract Document]
    D --> E[Email Metadata Analysis]
    E --> F[Document Content Analysis]
    F --> G[AI Classification]
    G --> H{Confidence >= 0.7?}
    H -->|Yes| I[Assign Document Type]
    H -->|No| J[Human Review Queue]
    I --> K[Route to Extractor]

Metadata Analysis

The classifier first analyzes email metadata:

// Quick classification from email only
const quickResult = classificationService.classifyByEmailMetadata({
  subject: "Invoice #12345 from Acme Corp",
  senderEmail: "billing@acme.com",
  hasAttachments: true
});
// Returns: { documentType: 'vendor_bill', confidence: 0.6 }

Pattern Matching:

Content Analysis

For higher confidence, the AI analyzes document content:

const result = await classificationService.classify({
  document: pdfBuffer,
  mimeType: 'application/pdf',
  emailSubject: subject,
  emailBody: body,
  senderEmail: sender,
  orgId: 'org_123'
});

Analysis includes:

• Document layout and structure
• Header text and titles
• Key phrases and terminology
• Address blocks (Bill To vs Ship To)
• Amount patterns and totals
• Company logos and branding

Confidence Scoring

Distinguishing Similar Documents

Vendor Bill vs Customer Invoice

The key distinction is direction of money flow:

Classification Hints:

• Check which company's logo/letterhead appears
• Look at the "Bill To" address
• Identify who is requesting payment

Vendor Bill vs Purchase Order

Receipt vs Vendor Bill

Custom Document Types

For specialized documents, create custom extractors:

// Custom document type
const customExtractor = await fetch('/api/extractors', {
  method: 'POST',
  body: JSON.stringify({
    name: 'Contract Agreement',
    documentType: 'custom',
    fieldDefinitions: [
      { name: 'contractNumber', type: 'text', required: true },
      { name: 'parties', type: 'array', required: true },
      { name: 'effectiveDate', type: 'date', required: true },
      { name: 'termLength', type: 'text', required: false }
    ]
  })
});

Best Practices

Improving Classification Accuracy

1. Use clear email subjects: Include document type keywords
2. Maintain sender consistency: Consistent vendor email addresses
3. High-quality documents: Clear, readable scans
4. Configure overrides: Known senders with known document types

Handling Low Confidence

When classification confidence is low:

1. Document goes to review queue
2. User manually selects document type
3. System learns from corrections
4. Future similar documents classified better

Multi-Document Emails

If an email contains multiple document types:

// Each attachment is classified independently
{
  emailId: 'email_123',
  attachments: [
    { filename: 'invoice.pdf', documentType: 'vendor_bill', confidence: 0.92 },
    { filename: 'receipt.jpg', documentType: 'receipt', confidence: 0.88 }
  ]
}