How do I use an OCR API for invoice data extraction?

To use an OCR API for invoice data extraction, you generally follow this workflow:

1. Pick an OCR/invoice extraction API

Choose a service that supports:

• Text OCR: extracts raw text from the invoice
• Document parsing / field extraction: identifies structured fields like invoice number, date, vendor, total, tax, line items
• Tables/line items: important for invoices

Common options include:

• Google Document AI
• AWS Textract
• Azure Document Intelligence
• Mindee
• Veryfi
• ABBYY
• Open-source OCR + your own parsing

2. Send the invoice file

Most APIs accept:

• PDF
• PNG/JPG/TIFF
• Sometimes multi-page documents

Typical upload methods:

• multipart/form-data
• Base64-encoded file
• Cloud storage URL

3. Receive extracted text or structured JSON

Depending on the API, the response may include:

• Raw OCR text
• Bounding boxes for words/lines
• Prebuilt invoice fields
• Confidence scores
• Line items in table form

4. Map fields to your schema

Usually you want to normalize extracted data into your own structure, for example:

• invoice_number
• invoice_date
• vendor_name
• vendor_address
• subtotal
• tax
• total
• currency
• line_items[]

5. Validate the output

OCR is imperfect, so you should check:

• Missing fields
• Confidence thresholds
• Currency/date formats
• Totals matching subtotal + tax
• Duplicate invoices

---

Example: generic OCR API request

Here’s a simple example using curl with a file upload:

curl -X POST "https://api.example.com/v1/invoices/extract" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@invoice.pdf"

Example JSON response:

{
  "invoice_number": "INV-10021",
  "invoice_date": "2026-10-01",
  "vendor_name": "Acme Supplies Ltd.",
  "subtotal": 120.00,
  "tax": 12.00,
  "total": 132.00,
  "currency": "USD",
  "line_items": [
    {
      "description": "Printer paper",
      "quantity": 4,
      "unit_price": 20.00,
      "amount": 80.00
    },
    {
      "description": "Ink cartridges",
      "quantity": 2,
      "unit_price": 20.00,
      "amount": 40.00
    }
  ],
  "confidence": 0.98
}

---

Example in Python

import requests

url = "https://api.example.com/v1/invoices/extract"
headers = {
    "Authorization": "Bearer YOUR_API_KEY"
}

with open("invoice.pdf", "rb") as f:
    files = {"file": f}
    response = requests.post(url, headers=headers, files=files)

data = response.json()
print(data)

---

Best practices

• Use an API with invoice-specific extraction, not just plain OCR, if you need structured fields.
• Preprocess images for better results:
  • high resolution
  • correct orientation
  • clear scans
  • remove shadows and blur
• Keep a fallback for manual review when confidence is low.
• Log OCR outputs for debugging and model tuning.
• Use idempotency or deduplication if invoices can be re-uploaded.

---

If you want to build a pipeline

A typical invoice extraction pipeline looks like:

1. Upload invoice
2. OCR + field extraction
3. Normalize fields
4. Validate totals/date/vendor
5. Store results in database
6. Send low-confidence cases to human review

---

If you want, I can also show you:

• how to do this with Google Document AI, AWS Textract, or Azure
• a Node.js example
• a workflow for line-item extraction and validation