Invoice Processing Automation
Primary Infotech built an end-to-end document AI workflow that extracts information from PDF invoices, validates invoice data against purchase orders, and automatically routes approved information for processing.
- Faster Processing
- 90%
- Accuracy
- 99.5%
- Hours Saved / Month
- 200+
01
The Challenge
The finance team processed more than 2,000 invoices every month from 15 vendors — each using a different format. Manual data entry consumed over 200 hours a month and carried a 4.5% error rate.
Errors caused payment delays, strained vendor relationships and made month-end close a dreaded, all-hands exercise.
02
Our Solution
We designed a document-AI pipeline that combines OCR with LLM-based extraction, so any invoice layout is converted into a standard, structured record.
Each invoice is validated against purchase orders in the ERP. Matches are approved and routed automatically; discrepancies are flagged with a clear explanation for a human to review. The whole workflow is orchestrated in n8n with Supabase as the system of record.
03
Technology
n8n orchestrates ingestion from email and SFTP, OCR and LLM extraction, three-way matching against ERP purchase orders, and approval routing. Supabase stores every document, extraction and decision for auditability.
- OCR
- GPT-4o
- n8n
- Supabase
- PostgreSQL
- ERP Integration
04
The Results
Processing time per invoice fell from 12 minutes to 45 seconds. The error rate dropped from 4.5% to 0.5%, and more than 200 hours a month were redirected to strategic finance work.
Month-end close is now a routine review instead of a scramble.