How to extract data from an invoice PDF or image (4 methods)

Invoice data extraction sounds simple — vendor, dates, line items, total — but invoices from different suppliers look wildly different. The vendor name might be at the top or the bottom. Line items might be a table or a paragraph. Tax might be labeled 'VAT', 'GST', 'Tax', or nothing at all. Here are the methods that actually handle this variety.

Method 1: Copy-paste (works on tidy digital PDFs)

Open the invoice PDF in your browser, select the line-item table, paste into Excel. Works on about 40% of invoices where the PDF was generated by software (not scanned). Most of the time, the columns collapse or the amounts paste as text strings that Excel can't sum.

Method 2: Python with pdfplumber + regex

For a fixed supplier you process at volume — same ERP, same layout every time — a Python script pays off.

• pip install pdfplumber
• import pdfplumber, re
• with pdfplumber.open('invoice.pdf') as pdf:
• text = ' '.join(p.extract_text() or '' for p in pdf.pages)
• total = re.search(r'(?:Total|Amount Due)[:\s]+\$?([\d,]+\.\d{2})', text, re.I)
• if total: print(total.group(1))

The limitation: regex breaks the moment the supplier changes their template. 'Total', 'Grand Total', 'Amount Due', 'Invoice Total', 'Balance Due' all mean the same thing — you need a different pattern for each. For invoices from many suppliers, maintaining this becomes a project.

Method 3: Dedicated invoice parsing libraries

Invoice2data (Python, open-source) uses YAML templates — you define the fields and patterns per supplier once, and it extracts them on every future invoice from that supplier. Good fit if you have under 20 suppliers and stable layouts. Falls apart when a supplier updates their system.

Method 4: AI extraction (any invoice, any layout, no templates)

Multimodal AI reads invoices semantically — it understands that 'Bill To' means the customer regardless of position, that line-item tables always have quantity × unit price = amount, and that the total is the biggest dollar figure near the bottom. No template needed, no regex to maintain.

What you can extract from an invoice

• Vendor — company name, address, tax ID / VAT number
• Customer — bill-to name and address
• Invoice number and purchase order reference
• Issue date and payment due date
• Line items — description, quantity, unit price, line total
• Subtotal, tax (VAT/GST/HST), discounts, shipping
• Total amount and currency
• Payment terms and bank details (where printed)

Scanned vs digital invoices

Invoices received by email as PDF attachments are usually digital — they have a text layer. Invoices sent by post and scanned, or photos taken of paper invoices, have no text layer. pdfplumber and invoice2data return nothing on scanned invoices. AI extraction applies OCR automatically and handles both.

Processing invoices from many suppliers

For AP automation — extracting invoices from a high-volume accounts payable inbox — the practical setup is an API call per invoice. ExtractFox's API takes a file upload and returns the same JSON schema regardless of which supplier sent the invoice. You can write the AP system integration once and drop new suppliers without touching the code.

Invoice data extraction vs template-based parsers

Docparser, Nanonets, and similar tools expect a template or training set per document type or supplier. That works until a vendor redesigns their PDF — then someone rebuilds the parser. AI invoice data extraction skips templates: the model already knows what an invoice is, so the first PDF from a new supplier extracts the same fields as the hundredth. For teams with dozens of vendors, that's the difference between maintaining a parser library and uploading files.

Exporting to Excel, CSV, and ERP

Structured invoice output should land in accounting software without manual cleanup. Numbers must be numeric (not strings with currency symbols), dates ISO-formatted, and line items one row per item. Excel export gives finance a review step; JSON fits NetSuite, Xero, or a custom AP webhook. Map vendor → supplier_id, invoice_number → external_id, and line_items → purchase_invoice_lines once — the schema stays stable across suppliers.

Common invoice extraction mistakes

• Treating OCR text as structured data — you still need field detection on top of raw characters.
• Building regex per supplier — breaks on the first layout change.
• Dropping tax lines — VAT/GST often sits in a separate row from subtotal.
• Splitting multi-line descriptions into extra columns — keep the full description in one field.
• Ignoring currency — store currency once at header level, amounts as plain numbers.