Automating invoice processing for a regional logistics company
A finance team buried in supplier invoices now has an AI pipeline that reads, checks and posts them to the ERP — with people reviewing only the exceptions.
−80%
Invoice processing time
98%
Fields extracted correctly
3 wks
To first release
0
Extra hires needed
The challenge
What wasn't working.
Around 2,000 supplier invoices arrived every month by email, in dozens of layouts. Three people re-typed them into the ERP, matched them to purchase orders and chased missing details.
-
Days of delay before invoices were approved
-
Typing errors causing payment disputes
-
No visibility of what was stuck, or why
The solution
What we built.
We built an AI document pipeline that plugs into the existing inbox and ERP, so nobody had to change how they work.
-
Capture 01
Invoices are collected automatically from the shared inbox.
-
Extract 02
An AI model reads 14 key fields and matches each invoice to its purchase order.
-
Review 03
Anything uncertain goes to a reviewer with the fields pre-filled.
-
Post 04
Approved invoices sync to the ERP with a full audit trail.
Built with
The technology.
- OpenAI
- Python
- AWS
- PostgreSQL
- Make
- Docker
“Month-end used to mean late nights. Now most invoices are posted before anyone's had their coffee.”
