Automated Data Extraction: Pull Invoice Fields from PDFs with n8n

Automated data extraction means using software to pull specific fields, such as an invoice number, date and total, out of documents and turn them into structured data, so nobody has to retype them. The method you need depends on the document. For clean, text-based PDFs, simple rules are enough. For scans and messy layouts you need OCR or AI. This guide compares the options, then builds a free n8n workflow that extracts invoice fields from PDFs and sends anything it cannot read to a person. We tested it on three PDFs plus a scan.

By the 360Automate Editorial Team · Last updated September 22, 2026 · More document automation guides

Four ways to extract data automatically

Method	How it works	Best for	Weakness
Rules (patterns)	Search the text for labels such as “Invoice No” and capture what follows	Text-based PDFs from a few known senders	Breaks when the wording or layout changes
Templates	Map fixed positions on a page to fields	Forms and documents with one exact layout	Needs one template per layout
OCR	Convert an image or scan into text first	Scanned or photographed documents	Errors on poor scans; still needs rules afterwards
AI / machine learning	A model reads the document and finds fields by meaning	Many different layouts, unstructured text	Cost, privacy and the need to check results

Most real systems combine them: OCR turns a scan into text, then rules or AI find the fields, then validation catches mistakes. The workflow below covers the rules-plus-validation part, which handles a surprising share of business documents at no cost.

Build a PDF invoice extractor in n8n

You need n8n and an SMTP account (we used Mailpit as a test inbox). Download the workflow (JSON) and import it. We tested with three PDFs generated for the purpose, like this one:

Step 1: Receive the document

Add a Webhook node (POST, path extract-invoice, respond using the Respond to Webhook node). It accepts a file upload in a form field called file. Your email tool, form or storage service can post documents to it.

Step 2: Read the PDF text

Add an Extract From File node, choose the PDF operation and set the binary property to file. It returns the document’s text. If a PDF is only a picture, the text comes back empty, and the workflow uses that to detect scans.

Step 3: Pull out the fields

Add a Code node that searches the text with patterns for the invoice number, date and total, and lists whatever it could not find:

const invoice_no = find(/invoice\s*(?:no\.?|number|#)\s*[:#]?\s*([A-Z0-9][A-Z0-9-]{2,})/i);
const totals = [...text.matchAll(
  /\b(?:total\s*due|amount\s*due|balance\s*due|grand\s*total|total)\s*[:]?\s*(?:USD|EUR|GBP)?\s*[$€£]?\s*([\d,]+\.\d{2})/gi
)].map(m => m[1]);
const total = totals.length ? Number(totals[totals.length - 1].replace(/,/g, '')) : null;

if (text.trim().length < 30) return [{ json: { status: 'needs_ocr', missing: ['all fields'] } }];

Three details make this reliable. The pattern starts with \b, so “Subtotal” is not mistaken for a total. The workflow takes the last match, because the grand total comes after subtotals and taxes. And the date is normalized to YYYY-MM-DD whether the invoice says “2026-09-14” or “September 3, 2026”.

Step 4: Respond and route

Respond to the sender with the extracted JSON, then add a Switch node with two outputs: All fields found goes to an email for accounts, and the fallback Needs review goes to a person with the fields it did find and a list of what is missing.

Step 5: Test with real files

curl -X POST http://localhost:5678/webhook/extract-invoice -F "file=@invoice-acme.pdf"

{"file_name":"invoice-acme.pdf","vendor":"Acme Supplies Ltd","invoice_no":"INV-48213",
 "date":"2026-09-14","total":1274.4,"currency":"USD","missing":[],"status":"extracted"}

Document	Result
Acme invoice (“Invoice No”, “Total Due”)	Extracted: INV-48213, 2026-09-14, USD 1,274.40
Northwind invoice (“INVOICE #”, “Amount due”, written date)	Extracted: NW-2091, 2026-09-03, USD 3,560.00
Harbor Print (no number, date or total)	Needs review: invoice number, date and total missing
Scanned invoice (image only)	Needs OCR: no text found

Pitfalls to expect

• Layout drift. A new supplier with different labels will fail. That is fine, as long as failures go to a person, and you add the new wording afterwards.
• Scans. A scan is a picture. Add an OCR step, from an OCR service or an AI vision model, before this workflow, and test it before trusting it.
• Losing the file after extraction. The Extract From File node returns text, so refer back to the original with $('Receive document').item.binary.file when you need the file name or want to save it.
• Guessing. Never fill a missing field with a default. A wrong total is worse than an empty one.

Measure accuracy before you trust it

Collect 20 to 50 real documents from your actual senders, run them through, and compare each field with the truth. Track accuracy per field, not just overall, because a workflow that gets dates right and totals wrong is not 90% correct in any useful sense. Re-test whenever a supplier changes its format, and keep a person reviewing exceptions.

Make it production-ready

• Send the fields somewhere useful: a spreadsheet, a database or your accounting system, rather than only email.
• Feed an approval flow. Extracted invoices can go straight into our invoice approval workflow.
• Prevent duplicates by checking the invoice number and vendor before saving.
• Secure the webhook with a secret header and limit file size and type.
• Protect the documents. Invoices and contracts contain personal and financial data, so control access and avoid sending them to services you have not vetted.

Frequently asked questions

What is automated data extraction?

It is the use of software to find and capture specific information from documents, emails or web pages and convert it into structured data such as a table row or JSON, without manual typing.

Can n8n extract data from PDFs?

Yes. The Extract From File node reads text from PDFs, and a Code node or an AI step can then find the fields. Scanned PDFs need OCR first.

What is the difference between OCR and data extraction?

OCR converts an image into text. Data extraction finds the specific fields, such as a total, inside that text. Scanned documents usually need both.

How accurate is automated data extraction?

It depends on the documents and the method. Clean, consistent PDFs can be near-perfect with rules, while scans and varied layouts need testing. Measure accuracy on your own documents and keep a human review step.

Sources and further reading

• n8n Extract From File node.
• n8n Webhook node.
• n8n Code node.
• Mailpit, the test inbox used for the screenshots.