Finance reconciles thousands of bank transactions to GL entries every month-end. Manual matching by ref + amount + date is tedious; close-cycle slips.
The AI approach
Match bank txns to GL entries by ref/amount/date with tolerance; LLM proposes candidate match for unstructured memos; rules auto-clear; exceptions to finance.
The outcome
~92% auto-cleared
Month-end close from 5 days → 2 days. Finance reviews only the 8% exceptions instead of every line.
Try itInput → Process → Output
Input — bank txns + GL entries
Bank txns · 4
2026-04-30 INV-7710−84,200
2026-04-29 SAL-6612+12,450
2026-04-28 TFR-INTL−5,820
2026-04-27 XFER-...A4−1,030
GL entries · 4
AP · INV-7710 Tipa V−84,200
AR · SAL-6612 Maxx+12,450
FX-loss adj−5,840
(unmatched)—
Process — AI pipeline
1Extract bank txns + GL entriesReadSymbolic
2Apply exact-match rulesRulesSymbolic
3Fuzzy-match remaining (memo)ReadGenerative
4Auto-clear or queue exceptionWriteSymbolic
Output — matched + exceptions
Click Run demo to apply the matching rules and clear what fits.
Match results 3 cleared · 1 exception
Bank txn
GL match
How
INV-7710 · 84,200
AP · INV-7710
exact ref
✓
SAL-6612 · 12,450
AR · SAL-6612
exact ref
✓
TFR-INTL · 5,820
FX-loss adj · 5,840
fuzzy + tolerance ±25
✓
XFER-...A4 · 1,030
(none in GL)
memo opaque
→ exception
Auto-cleared 3/4 (75% on this sample · ~92% across full month). Exception XFER-...A4 queued for finance — likely a personal transfer mis-routed to the company account; LLM suggests checking with bank.
Three AI types in this use case
SymbolicMatching rules (ref equality; amount within ±1฿; date ±3 days); GL coding rules; auto-clear thresholds; FX adjustment policies.
PredictiveString similarity / fuzzy match scoring on memos; memo categoriser (transfer / fee / payment / refund).
GenerativeLLM proposes probable match for unstructured memos ("XFER-XXXX from MR.SOMCHAI" → match to invoice INV-7710 on date proximity).