The problem
Bank reconciliation is matching free text to records. The payer writes the reference wrong, abbreviates the company name, pays two invoices at once, or pays a slightly different amount. Exact-match rules clear the easy lines and leave the rest to a person.
How it works
A payment that quotes the reference of an open invoice is a lookup, and the demo matches it without Aito. For the rest, the bank transaction's description, amount, and payer name go into the where clause, and _predict of invoice_id follows the schema link to the invoices table, limited to the open ledger. The candidates come back ranked, and $why highlights the tokens that carried the match: a company name, a word in the description, the amount.
The badge on each match shows Aito's $p, the number its explanation multiplies out to. The demo then weighs that probability against a plain amount check to decide what to auto-match, suggest, or leave unmatched. Only the confident matches post without review.
For the full architecture, see the technology overview. For the broader narrative across multiple use cases, read The Predictive Application.
See it live
This use case runs in the 🧾 Accounting demo today. Click through to the live application and inspect the queries that produce the result.
