As of 13 August 2026, AI can match bank payments to invoices.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, power-user skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Export the relevant bank transactions as a CSV or spreadsheet, keeping the transaction ID, date, amount, currency, payer name and payment reference.
   2. Export the open and recently paid invoice ledger, keeping the invoice number, customer name, invoice date, due date, original amount, outstanding balance and currency.
   3. Remove unnecessary bank account details and redact full account numbers before pasting either file into a chatbot or uploading it to an AI tool.
   4. Paste the prompt followed by the bank transaction export and invoice ledger, then ask the model to produce the confirmed, probable, ambiguous and unmatched sections.
   5. Compare every confirmed match against the original bank export and invoice ledger, checking the reference, customer, amount, currency and remaining balance.
   6. Open the ambiguous and partial-payment items in your accounting system and decide the correct allocation, fee treatment, credit note treatment or customer query.
   7. Post only the matches you have checked, then compare the resulting customer balances and total bank receipts with the AI report.

   Prompt

   Match the bank payments in the transaction table below to the invoices in the invoice table.
   
   Use these rules:
   1. Match on the strongest available evidence, including invoice number in the payment reference, exact or near-exact amount, customer name, currency and timing.
   2. Do not invent invoice numbers, customers, amounts or explanations.
   3. Treat each payment and invoice as one-to-one unless the data clearly shows a partial payment, combined payment, credit note, bank fee, refund or overpayment.
   4. Separate confirmed matches from probable matches, ambiguous matches and unmatched items.
   5. Do not force a match. Flag any payment that could match more than one invoice and any invoice that appears to have been paid more than once.
   6. For each proposed match, show payment ID, invoice number, payment amount, invoice amount, customer, match status, confidence as high, medium or low, and the evidence used.
   7. For partial payments, combined payments, fees, refunds and overpayments, explain the remaining balance or difference without deciding how it should be posted.
   8. Finish with totals for matched payments, unmatched payments and invoices still outstanding. Check that the totals reconcile to the input tables.
   
   This is an operational matching report, not accounting or tax advice. I will make the final posting decisions.
   
   Bank transactions:
   [PASTE BANK TRANSACTION EXPORT HERE]
   
   Invoice ledger:
   [PASTE INVOICE LEDGER HERE]

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.