Invoice Extraction Workflow

Automated extraction of supplier invoice data from bulk-scanned PDF documents using Claude AI Vision.

System: Invoice Extraction Server (Node.js + Claude API) Integration: Zoho Creator → Extraction Server → Zoho Creator Processing time: ~2-5 minutes per bulk PDF (depends on page count)

Overview

The invoice extraction workflow processes bulk-scanned PDFs from Australian auto parts suppliers, automatically splitting them into individual invoices and extracting structured data for matching against purchase orders.

What It Extracts

Field	Description	Example
`supplier_name`	Supplier/vendor name	"Australian Automotive Group"
`invoice_number`	Invoice identifier	"INV-2024-12345"
`customer_account`	CrashClaim's account number	"ACC12345"
`po_number`	Purchase order reference	"PO-2024-001"
`invoice_date`	Invoice date	"15/07/2026"
`vehicle_id` / `rego`	Vehicle registration	"ABC123"
`sub_total`	Subtotal excluding GST	450.50
`invoice_total`	Total including GST	495.55
`freight`	Freight/delivery charge	12.00
`line_items[]`	Array of invoice line items	See below

Line Item Structure:

description - Part description
part_number - Manufacturer part number
quantity - Quantity ordered
unit_list - List price ex GST
unit_net - Net price ex GST (expected)
total - Line total ex GST

Process Flow

graph TD
    A[User uploads bulk PDF to Zoho] --> B[Zoho Creator webhook triggers]
    B --> C[Extraction Server: Download PDF]
    C --> D[Split PDF into pages]
    D --> E[Detect blank pages]
    E --> F[Group pages into invoices]
    F --> G[Extract data with Claude Vision]
    G --> H[Validate extracted data]
    H --> I[Send data to Zoho API]
    I --> J[Zoho creates/updates records]
    J --> K[Send push notification]
    K --> L[Complete]

Step-by-Step Process

1. File Upload & Trigger

Who: Parts team, Claims handlers Where: Zoho Creator → Upload Parts/Order Invoice report What happens:

User uploads bulk-scanned PDF containing multiple invoices
Zoho Creator webhook fires immediately
Webhook payload includes:
- Record ID
- File URL
- Environment (dev/production)

2. PDF Download

System: Invoice Extraction Server What happens:

Server downloads PDF from Zoho Creator using OAuth token
Saves to temporary directory: temp/{job_id}/
Retry mechanism: 3 attempts with 3-second intervals

:::info Retry Logic If the download fails (network error, timeout, etc.), the server automatically retries up to 3 times with a 3-second delay between attempts. This handles transient network issues without manual intervention. :::

3. PDF Processing

Technology: pdf2pic + GraphicsMagick What happens:

PDF split into individual page images (PNG format)
Each page converted to high-resolution image for Vision AI
Blank pages detected using pixel analysis
Output: Array of page images in temp/{job_id}/pages/

4. Invoice Grouping

Algorithm: Sequential page analysis What happens:

Analyzes page patterns to identify invoice boundaries
Groups consecutive pages belonging to same invoice
Tracks page ranges: {invoiceNumber: 1, startPage: 1, endPage: 3}

5. Data Extraction (Claude Vision)

Model: Claude Opus 4 (Anthropic) What happens:

Each invoice (group of pages) sent to Claude Vision API
AI extracts all header fields and line items
Returns structured JSON matching schema
No retry mechanism - single attempt per invoice

Prompt engineering:

Supplier-specific extraction rules
Australian invoice format handling
GST calculation validation
Part number normalization

6. Data Validation

Checks performed:

Required fields present (supplier, invoice number, total)
Date format validation (dd/MM/yyyy)
Numeric field validation (totals, quantities)
GST calculation verification (10% of subtotal)
Line item total = sum of all line totals

7. Send to Zoho API

Endpoint: Custom Zoho Creator API Environment-aware:

Development: dev_invoice_extraction
Production: Invoice_Extraction_Integration

Payload:

{
  "record_id": "1234567890",
  "invoice_number": "Invoice_1",
  "page_count": 3,
  "start_page": 1,
  "end_page": 3,
  "extraction_success": true,
  "supplier_name": "Australian Automotive Group",
  "customer_account": "ACC12345",
  "po_number": "PO-2024-001",
  "invoice_date": "15/07/2026",
  "freight": 12.00,
  "vehicle_id": "ABC123",
  "rego": "ABC123",
  "sub_total": 450.50,
  "invoice_total": 495.55,
  "line_items": [...]
}

No retry mechanism - single attempt

8. Zoho Record Creation

What happens in Zoho:

Creates individual invoice record per extracted invoice
Links to original bulk upload record
Stores extracted data in structured fields
Updates bulk upload status to "completed"

9. Push Notification

Trigger: After all invoices processed Who gets notified: Parts team Channel: Zoho notification system

Error Handling

Error Type	Handling
File download failure	Retry 3 times with 3s delay, then fail job
PDF processing error	Mark invoice as failed, continue with others
Claude API error	Mark invoice as failed, log error, no retry
Zoho API error	Mark invoice as failed, log error, no retry
Validation failure	Mark as failed with validation errors

:::caution No Automatic Retries for AI/API Calls Claude API calls and Zoho API calls do NOT have automatic retry mechanisms. If these fail, the invoice is marked as failed and requires manual review. :::

Environment Configuration

The system supports two environments:

Environment	Zoho API	Use Case
Development	`dev_invoice_extraction`	Testing, UAT
Production	`Invoice_Extraction_Integration`	Live processing

Environment is specified in the webhook payload.

Monitoring & Status Tracking

Job Status Endpoint: GET /api/jobs/:job_id

Status values:

processing - Extraction in progress
completed - All invoices processed successfully
failed - Critical error occurred

Progress tracking:

Total invoices detected
Invoices processed so far
Invoices failed
Current step description

Temporary File Cleanup

When: After job completion (success or failure) What: Deletes entire temp/{job_id}/ directory Why: Prevent disk space issues, remove sensitive data

Best Practices

For Parts Team

Scan quality matters - Clear, high-contrast scans extract better
One supplier per PDF - Better grouping accuracy
Remove blank pages - Faster processing
Check extraction results - Review failed invoices manually

For IT/Developers

Monitor logs - CloudWatch/server logs show extraction issues
Failed invoice review - Manual intervention needed for extraction failures
API rate limits - Claude API has rate limits, batch uploads carefully
Test in dev first - Always test bulk uploads in development environment

Troubleshooting

Common Issues

Issue	Cause	Solution
File download fails after 3 retries	Zoho API timeout, invalid URL	Check Zoho file URL, verify OAuth token
All invoices fail extraction	Claude API key invalid/expired	Verify `ANTHROPIC_API_KEY` in environment
Wrong invoice grouping	Poor scan quality, missing pages	Rescan with higher quality, include all pages
Missing line items	Low image resolution	Increase scan DPI to 300+
Zoho API errors	Wrong environment, invalid payload	Check environment setting, verify API keys

Technical Details

Server: Node.js 20 + Express Deployment: Docker container on EC2 Image: AWS ECR (invoice-extraction-server:latest) Dependencies:

@anthropic-ai/sdk - Claude AI API
pdf2pic - PDF to image conversion
sharp - Image processing
axios - HTTP client
pdf-lib - PDF manipulation

Resource requirements:

Memory: 2GB minimum
Disk: 10GB temporary storage
Network: Stable connection to Zoho + Anthropic APIs

PO PDF Extraction - Cost Price Report extraction
Parts Assessment Extraction - Parts & Sublet List extraction
Reference: Order Parts Form - Destination for matched line items
Reference: Purchase Order Form - PO matching rules

For technical support or issues, contact the development team.

Overview​

What It Extracts​

Process Flow​

Step-by-Step Process​

1. File Upload & Trigger​

2. PDF Download​

3. PDF Processing​

4. Invoice Grouping​

5. Data Extraction (Claude Vision)​

6. Data Validation​

7. Send to Zoho API​

8. Zoho Record Creation​

9. Push Notification​

Error Handling​

Environment Configuration​

Monitoring & Status Tracking​

Temporary File Cleanup​

Best Practices​

For Parts Team​

For IT/Developers​

Troubleshooting​

Common Issues​

Technical Details​

Related Documentation​