Intelligent OCR System for Invoices and Delivery Notes
Industrial distributor · 6 weeks
95%
Processing time reduction
<1%
Error rate (vs 5% manual)
100%
Document traceability
Weeks
Visible ROI
Context
The company is an industrial distribution and logistics operation receiving dozens of invoices and delivery notes daily from multiple suppliers, each with completely different formats.
The challenge
- Every supplier has a different invoice format (PDF, scanned, emails)
- Manual entry generated a 2-5% transcription error rate
- Price discrepancies detected weeks later
- Repetitive workload for the admin team
The solution
Automatic data extraction
The system uses computer vision models (Mistral OCR, Google Gemini) to identify and extract supplier, invoice number, date, line items, totals, VAT breakdown and withholding taxes.
Intelligent reconciliation engine
Compares individual line sums with declared totals. When discrepancies are detected, triggers automatic AI correction.
Anomaly detection
Identifies outlier values (e.g. a total 10x higher than expected) and flags them as incidents with automatic email notification.
ERP integration
Validated data is automatically inserted into SQL Server, both headers and line items.
Results
- Documents requiring manual entry are processed automatically in seconds
- Automatic validation catches discrepancies that went unnoticed
- Complete traceability: every document generates an audit trail
- Scalable: ready to grow with business volume
Business impact
| Aspect | Before | After |
|---|---|---|
| Time per invoice | 3-5 minutes manual | Seconds (automated) |
| Error rate | 2-5% | <1% |
| Discrepancy detection | Weeks later | Instant |
| Admin hours/week | 15-25h | <2h (exceptions only) |
For a company processing millions in annual purchases, even a small error rate translates to significant undetected discrepancies. The system catches them before they become losses.
Status: IN PRODUCTION