Invoice reconciliation for a 900-supplier finance team

A finance team paying more than 900 suppliers was reconciling invoices against purchase orders and deliveries by hand, line by line.

We built an AI agent that reads supplier invoices, matches them to orders and goods received, posts what it is sure about, and sends everything else to a person with the reason attached. The first version was live in 11 weeks.

Invoices arrived in every format: PDFs, scans and emails, each supplier laid out differently. The team checked every line against the purchase order and the delivery note before approving payment.

The work was slow and repetitive, mistakes slipped through at month end, and the finance director wanted AI that would be honest about what it could not do rather than another tool that over-promised.

We built a document AI pipeline that extracts lines, quantities, prices and VAT from each invoice and matches them three ways against the purchase order and goods received.

The agent only posts a match it is confident about. Anything unusual, such as a price change, short delivery or duplicate, goes to an exceptions queue with the reason and the source documents side by side, so finance reviews exceptions instead of every line. Every decision is logged for audit.