The client
A Spanish professional services firm processing 400–600 supplier invoices monthly, across dozens of subcontractors and service providers. 3-person accounting team handling invoice intake through booking.
The problem
Invoice processing was entirely manual and serial:
- Receive PDF invoice by email
- Open, extract key data (amount, date, line items, tax IDs)
- Match against purchase orders and service agreements
- Flag any discrepancies for review
- Book into the accounting system
A single invoice touched 2–3 people and took 2–3 days end-to-end. Errors in data entry caused rework. The team was constantly behind, especially at month-end.
What we built
An end-to-end invoice automation pipeline:
- Monitors the dedicated invoices inbox automatically
- Extracts structured data from PDFs using OCR + AI (not just template-based — handles any format)
- Validates extracted data against existing purchase orders and supplier records in the ERP
- Matched invoices auto-booked into the accounting system
- Exceptions (mismatches, missing POs, unusual amounts) flagged for human review with full context
- Weekly summary report to the finance lead
Tech stack: Python, PDF extraction (OCR + Claude for structured output), ERP REST API, email monitoring via IMAP.
The result
- Processing time: from 3 days to 4 hours average
- 85% of invoices auto-processed without human touch
- Remaining 15% flagged for review have full context — no “detective work” needed
- Zero data entry errors on auto-processed invoices (measured against manual sample)
- Month-end close accelerated by 3 business days
The accounting team shifted from data entry to higher-value finance work: supplier negotiations, forecast analysis, and process improvements.