Three-way matching that stops eating a finance week
Purchase orders, delivery notes and invoices matched silently overnight. Only the exceptions reach a human in the morning.
The problem
Your accounts payable clerk opens a spreadsheet of open purchase orders, then digs through a shared inbox for supplier invoices and a folder of scanned delivery notes. They manually compare quantities, unit prices, line totals and dates across three separate documents for every single transaction. A single typo or missing delivery note can stall payment for days, and the whole process resets every Monday morning.
What AI makes possible now
An automation ingests purchase orders from your ERP, delivery notes from your freight portal or email, and supplier invoices as searchable PDFs. It parses line items, quantities, unit prices and totals across all three documents, then performs the three-way match. Matched transactions are logged with a timestamp and routed for scheduled payment approval. Exceptions, such as a quantity variance or a missing delivery note, land in a single queue with the specific discrepancy highlighted, ready for a human review.
How it works
- A monitored inbox and ERP connector pull in new purchase orders, delivery notes and invoices each night.
- A parsing model extracts line-level data from each document, regardless of layout, language or currency.
- A matching engine compares quantities, unit prices and totals across the three documents using configurable tolerance thresholds.
- Matched sets are posted for payment run approval. Mismatches appear in a review queue with the variance flagged and source documents attached.
The first thirty days
In the first thirty days, the automation connects to one core supplier and your ERP. It performs the full match on that single supplier's transactions, giving your team a real exception queue and a log of auto-matched entries. This validates accuracy on your actual data before scaling to all suppliers.
How it earns
It saves the cost of a part-time accounts payable clerk or stops a full-time one from drowning in paper. For a business processing a few hundred purchase invoices a month, it frees fifteen to twenty hours of manual reconciliation every week, turning a backlog into a same-day review of only the problematic items.
Why now
Supplier invoice formats remain wildly inconsistent, but document parsing models can now handle unstructured layouts without template setup. You can automate the matching without demanding every supplier change their invoicing process.
First customers
Mid-market importers, wholesalers or manufacturers running a busy finance function with a small team. Companies where the same person does the matching, the chasing and the data entry.
The hard part
A poorly scanned delivery note with a smudged quantity can cause a false match if tolerance thresholds are set too loose. The human review step must catch these, so the queue design matters as much as the model accuracy.
Build this with us
We map your procure-to-pay flow, connect the documents and build the matching logic with your approval rules baked in. If a week of your finance calendar is disappearing into paper reconciliation, you can apply to work with us on this.