Reconciliation & Matching
Two windows, side by side, scrolling. Bank statement on the left, ledger on the right. Ninety percent of the lines are the same transaction wearing slightly different clothes: a reference number truncated here, a date shifted a day there, a payment batched on one side and split on the other. An accountant burns a morning confirming what was never in doubt, and the four genuine breaks — the ones that actually matter — get found at 4pm when attention is at its worst.
A Skynet agent inverts the day. It clears the obvious matches and hands you a short list of exceptions, each with what it found and what it thinks the treatment is. You start with the hard part while you are still sharp.
How it works
Connect both sides
Point the agent at the sources you reconcile — bank feeds, the ledger, the payment processor, the billing system. Unified memory means it sees the same transaction across systems even when the identifiers do not line up neatly.
Auto-clear the certain matches
Where amount, date, and reference agree within the tolerances you set, the agent matches and clears. You define the tolerances: how many days of timing drift, how much rounding, which reference formats count as equivalent.
Work the exceptions
For everything left, the agent does the digging — searching for split payments, partial settlements, fee deductions, and duplicates across the connected systems — and proposes a treatment for each with the evidence attached.
Approve before it posts
Proposed journal entries wait for you. Nothing hits the ledger until a person approves it, and the agent records who approved what, so the trail is intact when someone asks later.
Build it from a prompt
Set the tolerances and let it run.
The exception list is where the value is. Because the routine matching is already done, breaks get investigated the day they appear rather than at close, and the reconciliation stops being a monthly wall of scrolling. Every clear is grounded in a defined rule, so you can explain how any given match was made.