The finance team at a property management group spent the early days of every month reconciling bank statements against the ledger by eye. Most lines matched cleanly on amount and reference, but enough did not — a payment split across two entries, a reference typed wrong, a fee deducted at source — that the work could not be trusted to anyone in a hurry. It was careful, repetitive, end-of-cycle work that ate days the team did not have, and it always landed in the same crowded week.
The easy matches were never the problem. The team’s time went on the handful that didn’t line up.
The challenges we had to solve
- A wrongly confirmed match quietly corrupts the books, so the tool could not be allowed to match anything on its own.
- The hard cases — splits, partial payments, transposed references — were exactly the ones a naive rule would miss.
- The analysts needed to see both sides of a proposed pair and the reason for it, not just an answer.
- It had to fit the close process they already ran, not add a step to an already tight week.
How we approached it
The assistant proposes; the analyst disposes. For each unreconciled line it suggests the most likely ledger entry — or the most likely combination, where a payment is split — weighing amount, date proximity, reference and counterparty, and it shows both sides side by side with the reason it thinks they belong together. The analyst confirms or rejects each suggestion. Nothing is posted as matched until a person says so. For the clean, unambiguous majority that is a fast confirmation; the team’s real attention goes to the awkward few, which is where it should have been all along.
We fed confirmations and rejections back, so the suggestions improved at the patterns specific to this group’s payers. The assistant worked against the ledger and the close process the team already used. We measured against the time to close and against how much of the reconciliation a person could clear without manually hunting for matches — and we kept a close eye on rejected suggestions, since a tool that suggests confidently wrong pairs would waste more time than it saved.
Where it stands
The close still belongs to the team, and every match is still confirmed by an analyst — but most of the line-by-line hunting has gone. They move quickly through the obvious pairs and spend their judgement on the splits and the mismatches, which is the part that needed a person in the first place. The crowded week is no longer quite so crowded.