One Hotel Bill, Nineteen Bank Charges

Many people think reconciliation means matching one receipt to one bank line.

In a real business, it is often many-to-many: several documents against several payments, with exchange rates, instalments and pre-authorizations in between. Do it by hand once, and the evening is gone.

Let me use two real scenarios — plus one from a friend's trade.

Scenario one: one business trip, one hotel.

A colleague spent five days in Jeddah. When he came back, the finance desk had:

One hotel folio. Twenty-five loose receipts — the hotel café, the main restaurant, room service, the bakery downstairs. And in the bank feed, nineteen charges, all showing nothing but the hotel's name.

The awkward part: the receipts' merchant names don't match the bank lines, and even the tax numbers differ — the folio is issued by the hotel group, the outlet receipts by the restaurant operator; legally two different entities. The amounts are all odd change: eight charges of 47.02, four of 90.13, and one big 3,559.61 for the room.

How does a human reconcile this? Spread out 45 documents and guess which payments add up to which receipts.

AI works differently: it first recognizes that these 26 documents belong to the same stay and bundles them; then it matches only within the bundle — the room to the one big charge, room service to the 47s, restaurant bills to the 90s. Nineteen bank lines, every one accounted for.

Scenario two: one bill, two charges.

Another hotel, in Saigon this time. One folio — but the card was charged in two instalments. No single bank line equals the bill total; only the two together equal it (within FX tolerance).

A human's first reaction is "was I overcharged?" AI does subset-sum: find the group of candidate charges whose sum equals the bill, match them, case closed.

Scenario three: a ship agent's PDA and FDA.

Friends in ship agency know this pain well. Before the vessel even arrives, you issue a PDA — a pro-forma disbursement account — and collect a prepayment from the principal. While the ship is in port, port dues, pilotage and towage keep arriving, each on a different supplier's invoice. After departure, you issue the FDA — the final disbursement account — and settle the difference, refunding or collecting the balance.

One engagement, spanning weeks: one prepayment, one top-up, one refund, with a stack of supplier invoices in between. Which bank line belongs to which section of which account — a human has to figure it all out.

AI can assemble that whole "estimate — prepayment — actuals — final settlement" chain into one complete engagement: who overpaid, who owes what, and how much.

Notice the pattern?

The hard part of bookkeeping was never reading a document. It is assembling evidence scattered everywhere into one complete transaction — even when it comes from different entities, different currencies, different weeks.

That used to be human work. Now AI does it: it looks at all the documents and all the bank lines together and finds the globally most sensible assembly — not first-come-first-served, match-at-random.

And whatever gets assembled automatically becomes audit-ready: receipt, bank line, evidence, all linked. Whatever doesn't fit, AI won't force — it is set aside for your judgment.

Human time goes only where human judgment is truly needed. The rest — go think about the business.