Ingest the statement and the book, match highest-certainty first — exact, tolerance, narration, many-to-one, one-to-many, then fuzzy — clear the exceptions, sign off under four eyes and publish the BRS. Every match carries its basis and its confidence. No black-box ML: a deterministic match is never replaced by a probabilistic one.
Tools that ‘learn’ matches produce a matched percentage and no reason. The unmatched residue is where fraud and error hide — and it is reconciled in a spreadsheet anyway.
A match without a basis cannot be defended when it turns out wrong.
Gateway settlements, sweeps and virtual accounts break 1:1 matching.
The statement that goes to the auditor is rebuilt outside the tool.
Levels are data — each with a match basis, tolerances, cardinality and a confidence band — so the rule that matched is always on record.
A matched item leaves the pool before lower levels run.
Gateway, sweep and virtual-account settlements resolved through narration and reference.
Unmatched items classified so each has an owner and an action.
Bank charges, interest and returns proposed as balanced journals for review.
Ingest, match, clear, sign, publish.
Matching stays rule-based and explainable. AI is aimed at the residue.
Unmatched items grouped and explained, with a proposed rule for a person to accept.
Across every FinIFRS product the rule is the same: a deterministic, tested engine computes and stores every figure first. AI reads documents, drafts sentences and answers questions over what is already stored — it never computes, posts or approves an amount.
Per-tenant rule sets; defaults when none are configured.
Against ML auto-match tools and the spreadsheet.
| Capability | ML auto-match | Spreadsheet | FinIFRS |
|---|---|---|---|
| Basis of every match on record | — | Manual | ✓ Level + confidence |
| N:1 and 1:N matching | Partial | Manual | ✓ Native |
| BRS published from the tool | Partial | Manual | ✓ Four-eyes |
Because every match must be explainable to an auditor. The cascade is deterministic and its string similarity is implemented in code, so the result is the same every run.
We reconcile a month of your busiest account and show you the basis of every match and the exceptions that remain.