An audit-ready walk from opening to closing — that explains itself
Turns a raw ledger extract into an Opening → Movement → Closing reconciliation and a cash bridge, with an AI layer that explains every material swing.
Context
Reconciling control accounts and explaining the period's cash movement was a manual, estimate-driven exercise. It was slow, hard to audit, and — because it leaned on memory and manual mapping — quietly error-prone: a ledger could be left out of the pull and no one would notice until something didn't tie.
Challenge
Replace manual estimates with a ground-truth, repeatable process that walks any group of ledgers from opening balance through every movement to closing balance, builds the cash bridge underneath it, and stands up to audit scrutiny.
What I built
A reconciliation engine that ingests a raw GL line-item extract and produces, for any group of accounts:
- An Opening → Movement → Closing walk, with movements broken out by document type and driver.
- A cash-flow bridge built from the same source data, so the cash story and the ledger story reconcile to each other rather than being assembled separately.
- An anomaly / variance layer powered by an LLM that explains every material swing in plain English — "this account moved because of X" — instead of leaving a reviewer to reverse-engineer it.
Because it reads source line items rather than summarised balances, it surfaces the gaps manual processes hide — for example, a ledger that was simply missing from an earlier pull. I packaged it as a reusable skill so the same walk runs every month with no rebuild.
Outcome
A monthly reconciliation that is fast, audit-ready, and self-documenting: numbers tie from source, every material movement carries its own explanation, and nothing depends on whoever remembers how it was done last time.