What is document classification?
Document classification is the task of deciding what kind of document a file is, such as an invoice, a contract, a receipt, a letter or a report. Once each file has a label, it can be sorted into the right folder, sent to the right person or processed with the right rules, without anyone opening it first.
How classification works
An AI model reads the text of the document and looks at its vocabulary, its structure and the phrases typical of each type. An invoice talks about amounts, due dates and tax, while a contract talks about parties, terms and signatures. The model then chooses the label that fits best. It works on the text layer, so a scanned file must go through OCR first.
Classification and data extraction
The two tasks are related but different. Classification answers the question of what the document is, while extraction pulls out facts, such as people, companies, dates and amounts, and returns them as structured data you can place in a spreadsheet. A common workflow does both: classify the incoming pile, then extract the key fields from the invoices.
Limits and good practice
Classification is a judgment, and mixed or unusual documents can be labeled wrongly, for example a contract with an invoice attached. Treat the label as a suggestion and spot check the results, especially when they trigger an action such as a payment. Extracted values should be compared with the document. Avoid sending documents with sensitive data unless you are comfortable with the service handling it.
Tools that use this
Guides that put it into practice
- Invoices for your accountantStep by step: merge a month of PDF invoices into one file, stamp a reference on every page, turn the bank statement into Excel and send it all at an e-mail-friendly size.
- Digitize a paper archiveStep by step: turn scanner and fax TIFF files into PDF, join each batch into one document, remove the blank backs of the sheets and shrink the scans for storage.
Common questions
- Does classification read the whole document?
- It reads the text of the file, which is enough in most cases. Long documents may be analyzed from the part that best shows their type.
- What if the label is wrong?
- Correct it and move on. Classification gives a fast first sort, not a guarantee, so keep a manual check for important files.
- Is extracting data the same as classifying?
- No. Classifying names the type of document. Extracting returns specific facts from inside it, such as dates, names and amounts.