How matching works

Member+

Once a statement is extracted, Inbox Ledger pairs each transaction with the invoice it most likely pays. It does this in two stages so the AI matcher only ever looks at a short, sensible list of candidates.

Stage 1: narrow the candidates

The first stage is a fast prefilter that runs over your unpaid invoices. For each transaction, it keeps only the invoices that pass all of these tests:

• Currency. The invoice currency matches the transaction currency.
• Date. The invoice falls inside a window around the transaction, from about one week before the invoice date to about two months after it.
• Amount. The invoice total is within about ten percent of the transaction amount.

An invoice that fails any test is dropped, which keeps the candidate list small before any AI runs.

Stage 2: score with an AI matcher

The surviving candidates go to an AI matcher. It reads the transaction details and the candidate invoices and gives each pairing a confidence score between zero and one.

A score at or above 0.6 becomes a suggested match. Anything below that threshold is discarded, so a weak guess never reaches your review queue.

Why two stages

Running the AI matcher against every invoice would be slow and costly. The prefilter does the cheap, mechanical work first, so the AI only weighs in on pairings that already agree on date and amount.

Related

Ready to try this?

Inbox Ledger turns your inbox into clean accounting data. The free tier includes 10 credits, refilled every 30 days.

Start free