The context
A building materials distributor sits between manufacturers and the construction trade. Every day it places purchase orders with suppliers, receives goods against delivery notes, and is invoiced for them. Before an invoice can be paid, someone has to confirm that what was ordered, what was delivered and what was invoiced all agree: the classic three-way match.
In construction supply, that is harder than it sounds. Orders are amended, deliveries arrive in parts, prices change, and documents come from dozens of suppliers in as many formats.
The challenge
Reconciliation was done by hand. A team of nine people spent their days opening documents, finding the matching order and delivery, and comparing them line by line before an invoice could be approved or a discrepancy raised with the supplier.
The work was slow and repetitive, and it scaled only by hiring. It was also where money leaked: when a team is working through a backlog, small overcharges, duplicate invoices and missing credits are easy to miss.
What we did
We built a system that does the reading and matching, and leaves people to handle the exceptions.
First, the system reads each incoming document, whatever its format, and extracts the information that matters: supplier, references, products, quantities, prices and totals. Next, it finds the corresponding order and delivery and matches them line by line, allowing for the realities of the business such as partial deliveries and agreed tolerances. Finally, anything that does not match is not guessed at. It is routed to a person with the reason it failed: a price difference, a quantity short, a missing delivery.
We ran the system alongside the existing team before relying on it, so its matches could be checked against the team’s own work.
The result
97% of documents are now matched automatically. The work that occupied nine people is handled by one, who reviews the exceptions the system flags. That is a reduction of almost 90% in manual reconciliation work, and discrepancies are now caught consistently before invoices are paid, rather than when someone happens to notice.
Why it worked
- It automated the routine and escalated the rest. The system never forces a match. Clear cases flow through; unclear ones reach a person with the reason attached.
- It fitted the documents the business actually receives. No supplier had to change how it invoices.
- It was checked before it was trusted. Running side by side with the team proved the match rate on real documents.
The same approach applies to supplier invoices, customer remittances, bank statements and freight bills. See our finance operations practice, and our article on where AI pays back in the back office.