Document Classification (Document module → Classify) organises documents against the labels that matter to your business: invoice vs purchase order vs contract; HR vs Finance vs Engineering; urgent vs routine. AI reads each document and assigns the best-matching category, at volumes and speed no manual sorting can match.
How it works
- Define your scheme — a list of categories with short descriptions (the clearer the definition, the better the results).
- Upload documents — individual files or a processed folder.
- Run — each document is read (parsing/OCR as needed) and classified, typically with a confidence level.
- Review — browse documents grouped by assigned label; adjust any the model got wrong, especially low-confidence items.
- Act — export the classification results or continue into a workflow that files, routes, or further processes each document by category.
Example schemes
| Scheme | Categories |
|---|---|
| Document type | Invoice, Receipt, Contract, Purchase Order, Report, Letter |
| Department | Finance, HR, Engineering, Legal, Operations |
| Sensitivity | Public, Internal, Confidential, Restricted |
| Urgency | Urgent, High, Normal, Low |
You can run multiple schemes sequentially — classify by type, then by department — to build a full taxonomy.
Making classification accurate
- Write category descriptions, not just names. "Finance: invoices, receipts, purchase orders and payment records" beats "Finance".
- Keep categories distinct. If two labels overlap ("Report" and "Analysis"), merge them or sharpen the wording.
- Provide an "Unclassified/Other" escape hatch so forced wrong-answers don't pollute real categories.
- Spot-check low-confidence results and feed corrections back by refining category wording.
- Pilot on samples first, then scale.
Where classification fits in automation
Classification is usually the first step of a document pipeline:
Upload → Classify (invoice?) → Extract (fields) → Approve → File/Post
Chain it with Data Extraction and approval steps in a Workflow for end-to-end processing. See the File Parser example solution.
Related
- Document Parser · Data Extraction · Document History
- Documents (Admin) — engines and limits