Invoice & Accounts Payable Automation

The Problem

A regional logistics company with operations across four states was processing over 1,200 vendor invoices per month. The accounts payable team of three clerks was doing every step manually: downloading PDF invoices from email, re-keying line items into a spreadsheet, matching against purchase orders in their ERP, deciding on cost centre codes, emailing department heads for approval, and finally posting to SAP.

Average processing time per invoice was 6 business days end-to-end. Processing cost per invoice — including labour, error correction, and late payment penalties — worked out to $18. The team was making 3–5 data entry errors per day, some of which triggered late payment penalties from vendors. Over Q3, these penalties totalled $12,000.

The Automation We Built

We designed a six-stage intelligent invoice pipeline:

1. Capture: A dedicated accounts-payable email inbox is monitored by n8n. Every new email attachment is extracted and classified — invoice, statement, or other.
2. AI extraction: GPT-4o Vision reads each invoice PDF and extracts structured data: vendor name, invoice number, date, line items, amounts, VAT, and payment terms. Extraction accuracy on this client's invoice corpus: 98.7%.
3. PO matching: The extracted data is matched against open purchase orders in SAP via API. Matched invoices (within a 2% tolerance) proceed automatically. Mismatches are flagged for human review.
4. Cost centre coding: An AI model trained on 18 months of historical coding decisions assigns the correct cost centre code for each line item. Accuracy on the validation set: 96%.
5. Approval routing: Invoices above threshold amounts are posted to a Slack channel as an approval card. Approvers click one button — approve or query. Approved invoices post automatically to SAP within 60 seconds.
6. Exception handling: Invoices that fail any validation step are queued in a human review dashboard with the specific failure reason highlighted, reducing resolution time from hours to minutes.

Results After 3 Months

95% of invoices now complete the full process from email receipt to ERP posting without any human involvement. Average processing time dropped from 6 business days to 4 hours. Per-invoice processing cost fell from $18 to $4.90 — a 73% reduction. Late payment penalties dropped to zero in the first full quarter of operation.

The three AP clerks were reassigned to vendor relationship management and strategic cash flow forecasting — higher-value work the company previously had no capacity for.

Tools Used

• n8n — pipeline orchestration and email monitoring
• GPT-4o Vision — PDF data extraction and cost centre coding
• SAP API — PO matching and ERP posting
• Slack — in-flow approval workflow
• PostgreSQL — audit trail and exception queue