As of 13 August 2026, AI can only partly reconcile customer payments against 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, chat-fluent 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, including transaction date, amount, currency, payer name, payment reference and a unique transaction ID.
   2. Export the open and paid invoice list from your accounting system, including invoice number, customer, invoice date, due date, gross amount, currency, credit notes and payment status.
   3. Gather remittance advices, credit notes, refund records and any existing payment allocation report for the same period.
   4. Remove unrelated transactions and redact bank account details or other personal data that the reconciliation does not need, then give the payment and invoice files clear names.
   5. Paste the supplied prompt into a chatbot and attach the payment file, invoice file and supporting records, replacing the bracketed file names with the actual names.
   6. Compare every confirmed match with the bank transaction and invoice, and inspect the possible-match, combined-payment, part-payment and duplicate lists against your accounting system.
   7. Resolve each exception using the remittance advice or customer record, then post or approve the allocations in your accounting system and retain the AI report as a working paper.

   Prompt

   Reconcile the customer payments in [BANK_PAYMENTS_FILE] against the invoices in [INVOICES_FILE]. These are business records, so do not invent, alter or round any values. Treat dates as UK dates in the format DD/MM/YYYY unless the file clearly states otherwise, and keep all amounts to the precision supplied. Use invoice number as the primary match, then compare customer name, payment reference, amount and date. Allow for exact matches, part-payments, overpayments, combined payments, credit notes, refunds, duplicate transactions and payments with missing references. Do not force a match when the evidence is insufficient. Return four tables: 1) confirmed matches with payment ID, invoice number, customer, payment amount, invoice amount and reason; 2) possible matches requiring my decision, with the competing invoices and the evidence for each; 3) unmatched payments; 4) unpaid or partly paid invoices. Add a short exception summary. Flag duplicate payments and any amount or currency mismatch. State which rows were excluded because the data was incomplete. Do not post entries, mark invoices as paid, give VAT advice or claim that a match is confirmed unless the supplied records support it.

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