Guide · Matching
Reconciliation matching rules: exact, tolerance and fuzzy narration, and when to use each.
Auto-match rates are decided by rules, not software. Good rules match the obvious items safely and leave only real exceptions for people.
Three kinds of rule
| Rule | How it matches | Use it for | Risk |
|---|---|---|---|
| Exact | Same reference, amount and date | Clean, high-volume flows with shared IDs | Low |
| Tolerance | Amount within a set difference, or date within a set window | Charges, rounding, FX, posting lags | Medium: set limits carefully |
| Fuzzy narration | Similar references or narrations, after cleaning | Bank statements where narrations are truncated or reformatted | Higher: needs review and limits |
Order matters
- Run exact rules first and take those items out of the pool.
- Run tolerance rules next, narrowest tolerance first.
- Run fuzzy rules last, and send low-confidence matches to review.
- Handle one-to-many and many-to-one matches, such as a bulk settlement against many transactions, as their own rules.
Govern the rules
- Write every rule down; do not leave it in a macro one person understands.
- Version rules and record who approved each change.
- Use maker-checker before a new rule takes effect.
- Review match quality: a high match rate with wrong matches is worse than a lower one.
- Set tolerances with finance; confirm accounting treatment with your auditor.
How AutoRecon does it
AutoRecon is built for finance, operations and treasury teams at banks, NBFCs and payment companies. Switch, network, core, GL, bank and MT950 statements and ERP files are loaded as they land and mapped to one schema, with bad rows rejected and the reason given. UPI and IMPS, cards, NEFT/RTGS, nostro and vendor recons are matched daily on rules you write: exact, tolerance and fuzzy narration matching. Each break is classified, aged and given an owner; recurring breaks become a suggested rule, which takes effect only after maker-checker. The daily close is approved by maker and checker, and MIS, board packs and regulatory returns are drafted with every number traced to the query that produced it. It is read-only by default and writes nothing to your core unless you open the path.
Questions
What are reconciliation matching rules?
Rules that decide when two items from different sources are the same: exact matches on reference, amount and date; tolerance matches within a set difference; and fuzzy matches on similar narrations.
What is fuzzy matching in reconciliation?
Matching items whose references or narrations are similar but not identical, after cleaning; low-confidence matches should go to review.
In what order should matching rules run?
Exact first, then tolerance from narrowest to widest, then fuzzy, with each pass removing matched items from the pool.
See it on your own data. AutoRecon — Close the day before it starts. Book a 30-minute working session with an engineer.