Worked example · Hospitality finance

Hotel invoice reconciliation automation: a worked example.

An illustrative example, not a delivered hotel deployment. It applies the payment-matching approach from our Wise payment gateway to the way payments reach a hotel finance team, and shows which parts are worth automating and which a person should still check.

The problem

Matching payments to invoices by hand.

A hotel receives bank transfers from corporate accounts, travel agents, group organisers and individual guests. Each payer writes the reference their own way: an invoice number, a booking number, a guest name, or the word for invoice in their own language.

Someone in finance reads each payment, finds the matching invoice and marks it paid. Until they do, the payment sits unmatched, the invoice still looks open, and a reminder can go out for a bill that has already been paid.

This is a good first automation target because most of it follows rules. The interesting part is drawing the line between what a rule can safely do and what a person should still decide.

How the matching works

Five steps from bank transfer to paid invoice.

The same flow as our Wise payment gateway for a hosting business, applied to hotel invoices.

  • 01
    Payment arrives
    The bank sends a notification for each incoming transfer (a webhook, in the Wise gateway), or the matching service reads the bank feed on a schedule.
  • 02
    Reference is read
    Patterns pick the invoice number out of the reference in the formats payers actually use, such as Invoice 3530, INV-3530, Factuur 3530, Rechnung-3530 or Facture 3530. The Wise gateway reads English, Dutch, German and French formats.
  • 03
    Amount is checked
    The amount must match the open invoice within a small tolerance the finance team sets, to absorb bank charges. The Wise gateway uses plus or minus €0.50.
  • 04
    Invoice is marked paid
    The payment is applied in the accounting or property management system, the payer can be notified and every match is logged.
  • 05
    Everything else goes to a person
    Payments that fail any check land on an exceptions screen, where someone in finance assigns them to an invoice or sets them aside. Nothing is guessed.
Automate or check

What to automate, and what a person should still check.

The split we would propose to a hotel finance team before building anything. Your own rules may draw the line differently.

PaymentAutomate?Why
One transfer, a clear invoice reference, the exact amountAutomateA rule can match it reliably and the result is easy to check.
A clear reference, the amount short by a small bank chargeAutomate, within the agreed toleranceFinance sets the tolerance once; anything outside it goes to a person.
A reference in another language or formatAutomateOne pattern per format the hotel actually receives.
One transfer paying several invoices, such as a travel agent's monthly statementSuggest, then a person confirmsSplitting a payment across invoices is a judgement call.
A deposit or part-payment for a group bookingPersonWhat counts as paid depends on the contract terms.
No reference, or only a guest namePerson, with suggested matchesA language model can suggest likely invoices from the free text; a person decides.
A payment in a different currency from the invoicePerson, unless rules are agreedExchange differences need a finance decision.
Refunds, chargebacks and possible duplicatesPersonThe risk is double-matching or paying out twice.

Illustrative. This table is a proposed starting point for a discussion, not results from a hotel.

Where AI fits

Mostly plain code, with AI for the messy references.

Most of this workflow needs no AI. Reading a known reference format, checking an amount and marking an invoice paid are rules: cheaper, faster and easier to test than a model.

A language model earns its place on the free-text cases, such as a reference that only names a guest or a company. There it suggests candidate invoices for a person to confirm; it does not mark anything paid by itself.

This is the same trade-off we describe on AI workflow automation, and the reason we often recommend a plain workflow over an agent: see AI agents for business.

Before you buy anything

How to decide whether it is worth automating.

A short check a finance team can run on its own records first:

  • Count a month of incoming transfers, and how many already carry a usable invoice reference. Measure it rather than estimate it.
  • List the reference formats and languages that actually arrive.
  • Agree the amount tolerance, and who owns the exceptions screen.
  • Check that your accounting or property management system can receive a payment through an API or an import.
  • Start with the clean-reference payments only, measure how many match, then widen the rules.

If the numbers are unclear, this is exactly what the AI readiness assessment ranks, alongside your other manual workflows.

Related

The evidence behind this example.

01
Bank reconciliation case study
The Wise payment gateway this example is based on: webhook matching, multi-language reference parsing and a manual matching screen for edge cases.
02
AI workflow automation
How we decide which steps are plain code, which need a language model and which stay with a person.
03
AI readiness assessment
A two-week sprint that maps your operations and ranks where automation pays back, including the ones not worth doing.

Want to know if this fits your finance team?

Discovery calls are free and last 30 minutes, with someone who will actually do the work. If a simpler tool solves it, we will say so.

The button opens our enquiry form; we reply by email to agree a time for the call.

Coverage: Europe, Asia and remote. Email [email protected].