Matching: the right invoice on the right expense.
An incoming invoice only has value once attached to the expense it justifies. The platform confronts each read document with the original transaction, computes a score, and decides against explicit thresholds: automatic, review, or discarded.
Three destinies for every document
Automatic match
Very high score: merchant, date, amount and lines all consistent. The supporting document is attached without intervention, and the score trail is kept.
Human review
Intermediate score: an explainable gap in amount or date. An operator confirms or rejects, and their decision stays attached to the file.
Outright discard
Below the floor, the proposal is dropped: the invoice matches nothing safely. Better a pending file than a phantom supporting document.
Why thresholds, not a binary yes
A service-station invoice may legitimately differ from the amount authorised at the pump by a few cents; a toll may be invoiced in arrears. All-or-nothing matching would break on these real cases. Thresholds, by contrast, are tunable per perimeter, and every decision — automatic or human — is archived with its score. That is what makes the whole set defensible in a sample-based audit.
Frequently asked questions
What is the matching score based on?
On what actually identifies the expense: the merchant, the date, the amount and the product lines. Each dimension weighs into the final score, and the date gap is bounded by a configurable window.
Can a match happen automatically on the wrong invoice?
Not beyond the thresholds. Automatic validation only happens above a very demanding score; below a floor, the proposal is discarded; in between, a human always decides before anything is written.
Who corrects the product lines?
The platform proposes the correction when the lines read diverge from what the merchant type expects, and the final write goes through review. Corrections are attributed to a user, dated, never erased.
Upstream: receipt OCR. Downstream: the accounting export. All of it serving VAT recovery.