The duplicate invoice that gets caught is the one sent twice with the same number, and every system catches it. The duplicate that gets paid is the one that does not look like a duplicate: a reissued document with a new number after a query, a charge appearing on both an invoice and a statement, or something bought on a card and later invoiced.
The reissue after a query
A supplier corrects something and issues a new invoice with a new number, and the original is still in your system. Number-based detection sees two distinct invoices. Catching this needs a check on supplier, amount and approximate date, and it needs somebody to look at the flagged pairs, because occasionally two similar invoices are both genuine.
The charge that appears twice in different forms
A deposit paid manually and then included in the final invoice. A delivery charge on both an invoice and a monthly statement. These are not duplicate documents at all, so no detection rule sees them. The defence is recording deposits against the order and reconciling statements rather than paying from them.
The purchase paid by two routes
Bought on a card by somebody in a hurry and then invoiced to the accounts address. This is common, invisible to payables, and prevented only by reconciling card spend against invoices. It is one of the strongest arguments for a purchase-order threshold that is low enough to cover the spend people put on cards.
Questions people ask about duplicate invoice
How do we find historical duplicates?
Query the last year for the same supplier, similar amount and near date, and reconcile statements for your largest suppliers. Both usually find something.
Will a supplier refund a duplicate?
Almost always, usually as a credit against the next invoice. It is an accounting nuisance for them too.
Should detection block payment?
Block on exact supplier and invoice number. Warn on fuzzier matches so people investigate rather than learning to override.