Why this matters
Invoices, receipts, contracts and forms arrive as PDFs, and somebody retypes them: vendor, invoice number, date, total, renewal date. It is slow, and a single misplaced digit can mean paying the wrong amount or missing a deadline.
Document AI does the reading for you. Every tool in this course, from Smallpdf and KDAN to Docparser, Foxit and an n8n invoice workflow, relies on the same three steps: get the text out of the file, find the fields in that text, and check them before they count. Knowing those steps tells you which tool to pick, and where to look when the result is wrong.
What you’ll build
You follow two vendor invoices at Kestrel Design Studio (made up), one digital PDF and one phone scan, through OCR, parsing rules vs AI extraction, and the four checks that stop a misread total from reaching the books.
What you need
- Nothing but a browserIt is a short explainer of diagrams you click to open. No account, no files and nothing to set up.
Concepts
3 ideas you’ll use in the practice.
OCR turns a picture of a page into text
A PDF from billing software already has text inside. A scan or a phone photo is a picture, and OCR (optical character recognition) has to read it first. The result is often a searchable PDF: the same image with invisible text laid over it. Numbers on scans are where OCR slips, such as a zero read as the letter O.
Parsing finds the fields
Parsing rules take the text next to a label in a known layout ("whatever follows Invoice #"), which is exact and cheap when a vendor’s invoices always look the same. AI extraction asks a model for each field by name and works across layouts, as long as it is told to leave a field blank rather than guess.
Checks make the data safe to use
Do the lines add up to the total, is the invoice number a duplicate, is any required field blank, and is it over the approval limit? These four checks catch most extraction mistakes, and keep a person in charge of what gets paid.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Quick check
3 questions to check you’ve got it.
1. You cannot select any text in a PDF. What does it need before an AI tool can read it?
- A password
- OCR
- A digital signature
Show the answer
OCR. A scanned PDF is a picture of a page. OCR turns it into text that tools can read and search.
2. Every invoice from one vendor has the same layout. Which parsing method fits best?
- Parsing rules
- Retyping by hand
- Asking a chatbot to guess
Show the answer
Parsing rules. Rules that read the text next to a label are exact and cheap when the layout never changes.
3. The extractor cannot find a due date on an invoice. What is the right result?
- Fill in 30 days from today
- Leave the field blank and flag it for review
- Skip the invoice
Show the answer
Leave the field blank and flag it for review. A blank you can see beats a guess you cannot. A person fills it in from the vendor’s terms.
Smallpdf: Get the Key Points of a 25-Page PDF in 5 Minutes
You know how software reads a document. Next, use it: get the key points of a 25-page PDF with Smallpdf, and check every answer on the page.
Start module 3 →