Archevi
aiextractionocrprocessing

How AI Extraction Works

2 min read

When you upload a document to Archevi, our AI automatically analyzes it to extract key information. This makes your documents instantly searchable and enables features like expiry alerts.

The Extraction Process

Here's what happens when you upload a document:

  1. Text Recognition (OCR) - The AI reads all text in your document, even from photos and scans
  2. Document Classification - Identifies the document type (passport, receipt, contract, etc.)
  3. Entity Extraction - Finds and labels important information like names, dates, and numbers
  4. Relationship Mapping - Connects extracted data (e.g., "expiry date" belongs to "passport number")
  5. Indexing - Makes everything searchable for instant retrieval

What Gets Extracted

The AI looks for different information depending on document type:

Identity Documents

  • Full name, date of birth
  • Document number (passport #, license #)
  • Issue and expiry dates
  • Issuing authority

Financial Documents

  • Account numbers
  • Amounts, balances, totals
  • Transaction dates
  • Merchant/vendor names

Insurance Documents

  • Policy number
  • Coverage amounts and deductibles
  • Effective and renewal dates
  • Insured parties

Contracts and Legal

  • Party names
  • Effective dates and terms
  • Key obligations and amounts
  • Signatures and dates

Reviewing Extractions

After upload, you can review what the AI extracted:

  1. Open the document details
  2. Click the "Extracted Info" tab
  3. Review the labeled information
  4. Edit any incorrect extractions
  5. Add any information the AI missed

Improving Extraction Accuracy

For best results:

  • Use clear, high-resolution scans or photos
  • Avoid shadows, glare, and blurry images
  • Ensure all text is fully visible
  • Upload complete documents, not partial pages

See our supported documents guide for detailed tips on document quality.

Manual Corrections

The AI is highly accurate but not perfect. You can always correct extractions:

  • Click any extracted field to edit it
  • Add missing information manually
  • Remove incorrect extractions
  • Your corrections help improve future results

Privacy

All extraction happens on our secure Canadian servers. Your documents are never sent to third-party services. We don't use your data to train AI models. See our privacy documentation for details.