Bank reconciliation
Matching rules, confidence scoring, and how to train the reconciler on your own patterns.
Bank reconciliation in the platform is a mix of deterministic matching and machine-assisted suggestion. When a bank statement is imported — via file (CAMT.053, MT940, OFX, QIF, CSV) or direct feed — each transaction is evaluated against outstanding invoices, bills, and journal entries.
The matcher uses several signals: amount equality, reference-number match, counterparty fuzzy match, date proximity, and recurrence pattern. Each signal has a weight; the combined score determines whether the match is proposed automatically, queued for review, or left unmatched.
You can train the matcher on your own data. If you consistently categorise a particular vendor's monthly charge against a specific account, the matcher learns that pattern within three occurrences and starts suggesting it automatically.
Drag-to-match is supported for ambiguous cases. You can split a single bank line across multiple system transactions, or merge multiple bank lines into a single matched position for instalment payments.