← Back to case studies
Document AIAccountingPDF Parsing

From 3 hours of daily data entry to 9 minutes of exception review.

Freight forwarding company — 45 employees, Riga

Results

85%

of invoices processed without human touch

2.5 hrs

saved per day for the AP team

0.3%

error rate, down from 4%

3 weeks

from kick-off to production

The Challenge

The accounts payable team was manually keying 80–120 PDF invoices per day from 60+ suppliers into Microsoft Navision. Each invoice took an average of 3 minutes — longer when supplier formatting varied. Keying errors caused payment delays and strained supplier relationships. The team was spending roughly 15% of their total capacity on a task that produced zero business value.

The Solution

We built an AI agent that monitors a dedicated AP inbox around the clock. When a new invoice arrives, the agent extracts structured line items using a vision model, matches each line against the open purchase order database, and auto-posts matched invoices directly into Navision. Invoices that cannot be matched with high confidence are flagged in a daily exception digest for human review — a 9-minute task instead of a 3-hour one.

Tech Stack

GPT-4o VisionPythonMicrosoft Navision APIIMAP email monitoringPostgreSQL

Want similar results for your team?

Start a project