Automated Invoice Processing

The problem

Northwind’s finance team was processing ~600 invoices a month by hand. Each invoice took an average of 14 minutes, read it, type it into NetSuite, route it to the right approver, chase the approver. Friday afternoons were a queue of unprocessed PDFs.

The system

We built a pipeline that watches the shared inbox, extracts vendor, total, GL code, and PO match using an LLM extractor, and routes the invoice to the right approver in Slack with a one-click approve/reject. Edge cases (unmatched POs, weird vendors) escalate to a human dashboard instead of trying to fake it.

The result

After 6 weeks: 85% reduction in handling time, 99% accuracy on the matched PO field, and an unexpected win, the team finally had clean data to build month-end reporting against.

“I forget the system even exists. That’s the best review I can give it.” Sara Khan, Head of Finance