Extracting Invoice Data Using Regex and LLMs

The Invoice Data Entry Problem

Manual invoice entry is one of the most time-consuming bookkeeping tasks. Processing 100 invoices per month can consume 15-20 hours of manual data entry. With regex-powered AI, you can reduce this to under 1 hour while improving accuracy.

Critical Invoice Fields to Extract

Every invoice contains similar data points that regex can reliably identify:

1. Invoice Number Patterns

2. Date Extraction

Invoices use various date formats. Regex patterns for each:

Format	Pattern	Example
MM/DD/YYYY	\d{2}/\d{2}/\d{4}	11/15/2025
DD-MM-YYYY	\d{2}-\d{2}-\d{4}	15-11-2025
Month DD, YYYY	[A-Za-z]+ \d{1,2}, \d{4}	November 15, 2025
YYYY-MM-DD	\d{4}-\d{2}-\d{2}	2025-11-15

3. Amount Extraction

Currency amounts come in many formats:

• With currency symbol: \$[\d,]+\.\d{2} → $1,234.56
• Without symbol: \b\d+\.\d{2}\b → 1234.56
• With thousands separator: \$?[\d,]+\.\d{2} → $1,234.56 or 1,234.56
• Total line: (?i)total:?\s*\$?([\d,]+\.\d{2})

4. Vendor Information

Extract vendor details:

• Company name: Often in first few lines, all caps
• Address: Street, City, State ZIP pattern
• Tax ID: \d{2}-\d{7} (EIN format)
• Website: www\.[a-z0-9-]+\.(com|net|org)

AI + Regex Workflow for Invoices

Step-by-Step Process

1. Convert PDF to text

   Use OCR or PDF extraction tool (many AI platforms include this)

2. Apply regex pre-extraction

   Pull out obvious patterns: dates, amounts, invoice numbers

3. AI prompt with extracted data

   "From this invoice text, I've extracted:
   - Invoice number: [regex result]
   - Date: [regex result]  
   - Total: [regex result]
   
   Please verify these are correct and extract:
   1. Vendor name
   2. Billing address
   3. Line items with descriptions and amounts
   4. Tax amount
   5. Payment terms
   
   Return in JSON format."

4. AI processes and structures data

   Returns clean JSON with all invoice fields

5. Regex validation of AI output

   Verify amounts match pattern, dates are valid, invoice number format correct

6. Import to accounting system

   Direct API integration or CSV import

Advanced Extraction Patterns

Line Items

Extract individual line items from invoices:

Pattern: ^(.+?)\s+(\d+)\s+\$?([\d,]+\.\d{2})\s+\$?([\d,]+\.\d{2})$

Matches:
Office Supplies    5    $10.00    $50.00
Consulting Hours  10   $150.00  $1,500.00

Groups:
1. Description
2. Quantity
3. Unit price
4. Line total

Tax Amounts

Find sales tax or VAT:

Pattern: (?i)(sales?\s+tax|vat):?\s*\$?([\d,]+\.\d{2})

Matches:
Sales Tax: $45.67
VAT $123.45
Tax $5.00

Payment Terms

Pattern: (?i)(net|due in)\s+(\d+)\s+(days?|months?)

Matches:
Net 30 days
Due in 15 days
Net 60

Real-World Success Story

Best Practices

1. Vendor-Specific Patterns

Create custom patterns for your top 20 vendors—they likely represent 80% of invoice volume.

2. Validation Regex

After extraction, validate:

• Amount format is correct
• Date is not in future
• Invoice number is unique
• Subtotal + tax = total (AI can calculate, regex can validate format)

3. Confidence Scoring

Have AI rate extraction confidence. If <90%, flag for manual review.

Common Pitfalls to Avoid

• ❌ Making patterns too specific (won't catch variations)
• ❌ Making patterns too broad (false positives)
• ❌ Not testing on sample invoices first
• ❌ Forgetting case-insensitive matching
• ✅ Start with vendor-specific patterns
• ✅ Use AI to suggest patterns based on examples
• ✅ Validate extracted data before importing

Next in This Series

Continue learning:

• Date Format Standardization with Regex and AI
• Vendor Name Normalization Using Regex
• Receipt Parsing Automation with Regex