What this automation does
Every document that arrives at your organization — via email attachment, upload portal, scanner, or shared drive — gets read by AI, classified into a category, and routed to the appropriate destination. Invoices go to accounts payable. Contracts go to legal. Customer forms go to onboarding. Resumes go to HR.
In most businesses, a human manually sorts through incoming documents, reads enough to understand the type, and forwards or files them. This takes 5-15 seconds per document and is tedious work that leads to misfiling. AI classification is faster, more consistent, and scales without adding headcount.
Tools you need
- Gmail or shared inbox: Where documents arrive — email attachments, uploads, or scanner output
- OpenAI API or Claude: Reads document content and classifies by type ($0.02-0.05 per document)
- Make or Zapier: Routes classified documents to the right Google Drive folder, Slack channel, or team inbox
How to set it up
Step 1: Define your document categories. List every type of document your organization receives and where it should go: invoices to accounting, contracts to legal, purchase orders to procurement, resumes to HR, customer forms to onboarding, etc. Include 2-3 example descriptions for each category.
Step 2: Create a Make scenario triggered by new email attachments or new files in a shared drive folder. Extract the document text (for PDFs, use a text extraction step; for images, use OCR). Send the extracted text to OpenAI with your category list and examples.
Step 3: Ask the AI to return a JSON object with: document_type, confidence_score (0-100), a one-line summary, and the recommended destination (folder path, team, or email address). Based on the classification, route the document — move the file to the correct Google Drive folder, forward the email to the right team, or create a task in your project management tool.
Step 4: For the first two weeks, log all classifications and have a team member spot-check 10% of them. Adjust your category definitions and examples based on any misclassifications. Most teams reach 95%+ accuracy within the first week.
Cost breakdown
| Item | Cost | Notes |
|---|---|---|
| OpenAI API | $10-$20/mo | ~$0.03 per document at 300-700 documents/mo |
| Make or Zapier | $15-$25/mo | Based on document volume |
| OCR (if needed) | $0-$5/mo | For image-based documents |
| Setup time | 30-45 min | One-time |
| Total monthly | $25-$50/mo | Saves 4+ hours/week of manual sorting |
Frequently asked questions
The AI processes the full text content regardless of page count. For classification purposes, the first page is usually sufficient — invoices have header information, contracts start with party names and terms, resumes begin with contact info. The automation can be configured to read just the first page for speed or the full document for accuracy.
Yes. GPT-4 and Claude handle major languages well — English, Spanish, French, German, Portuguese, Chinese, Japanese, and more. The AI can classify documents regardless of language and even add a language tag to the output for additional routing logic.