Automated Email Invoice Processing

A client of Maxedv, an IT consulting company, handling large volumes of invoices via email needed to automate the capturing, extraction, and organization of these documents to improve operational efficiency.

The core idea behind the project was to:

• Automatically download invoice attachments received via email using an IMAP connection.
• Extract key invoice data reliably using AWS Textract OCR.
• Organize files locally, renaming them according to a strict predefined convention: [Owner]_[DocumentType]_[Provider]_[InvoiceNumber].[Extension].

From the beginning, the focus was on reducing manual time expenditure, minimizing data entry errors, and improving overall operational efficiency and cost-effectiveness.

The main challenge was to eliminate a highly inefficient and error-prone manual email and invoice processing workflow, while navigating complex technical hurdles in data extraction and system reliability.

Process Inefficiencies:

• Administrative employees manually opened emails, downloaded attachments, and entered data.
• The manual process had a high error susceptibility ("Fehleranfälligkeit") when sorting and renaming files.
• High time expenditure ("Zeitaufwand") directly affected costs and efficiency ("Kosten und Effizienz").

Technical Hurdles:

• IMAP State Management: Tracking the exact inbox state so the script wouldn't process the same emails twice, even after a system reboot or crash. Tracking total email counts proved unreliable.
• Data Extraction Consistency (The Regex Limitation): Initially, extracting the invoice number and vendor name relied on complex Regex and local Python libraries (like pdfplumber, EasyOCR, and Pytesseract). However, this approach struggled with varying invoice table formats, scanned images, and high CPU usage. Relying on the email sender's name for the vendor was also discarded as it required constant manual maintenance of a vendor list.
• Fuzzy Name Matching: Finding the exact internal owner ("Eigentümer") name on the invoice was difficult due to slight typos or variations in the document text compared to the internal database.