An anonymized account of preparing an inbound-payment review package. The documented result is a completed package for client review and submission; the subsequent bank outcome is not established.
Context
A Thai company had supplied invoices and incoming-transfer notices for several separate commercial transactions. It needed a usable way to ask its receiving bank what had happened to the payments and what, if anything, was still required.
One transaction exposed the central problem. The invoice and the bank notice showed different amounts. The notice also listed a charge equal to the difference. That made a fee deduction a plausible explanation, but did not establish it. The notice did not identify the invoice, and it explicitly said that it was not confirmation of payment.
Constraint
Those documents answered different questions. The invoice described the commercial amount. The notice recorded receipt of a payment instruction. Neither established that the company had been credited, and the relationship between the notice and the invoice still needed confirmation.
A single label such as “payment confirmed” would have hidden these distinctions. Changing an amount to make the documents agree would have removed the exception the bank needed to examine. Calling the difference an unpaid balance would have introduced another unsupported conclusion.
No separate contract had been supplied. I did not make obtaining one an automatic prerequisite for preparing this inquiry. The test was specific: what unresolved fact would that document establish for the question being put to the bank?
Decision
I set the structure around the company, each separate commercial transaction, and its associated payment records. Shared company information could be reused; different buyers and invoices could not be merged merely because they involved the same beneficiary account.
Within each transaction, the workflow had to distinguish what a document stated, what the client reported, what we inferred, and what remained unknown. The bank notice could support particular instruction details without proving final credit.
For the disputed transaction, I required both amounts to remain unchanged. The apparent connection to the charge was retained as a question to resolve, not promoted into an explanation.
Implementation
AI assisted with extracting the supplied documents, comparing the fields, drafting the inquiry and assembling the package under those instructions. I defined the evidence boundaries and reviewed whether the handoff was usable by the client. This was preparation for a bank decision; it did not transfer that decision to AI.
The resulting transaction folder contained the original invoice, the notice extracted from its PDF portfolio, and a separate document with the proposed email. The inquiry asked the bank to confirm the invoice linkage, the instructed amount, how the charge had been applied, and the current status. If credit had occurred, it asked for the net amount and credit date. If more evidence was required, it asked the bank to identify the specific gap.
The handoff also needed operational detail. During review, I identified that the recipient address was missing from the individual letter documents. I required that to be corrected and the client instructions simplified. The final version specified the recipient, subject, message and exact attachments. It told the client to preserve the source documents and retain the sent correspondence and full bank response.
Evidence
The retained invoice and bank notice support the discrepancy and the notice’s limitation. The final inquiry preserves both instead of presenting a resolved payment narrative. The packaged invoice matches the supplied original byte for byte; the packaged notice matches the document embedded in the original portfolio.
The original task and subsequent review messages also distinguish my contribution from the AI’s: I set the transaction structure, prohibited unsupported reconciliation, challenged unnecessary paperwork and required an executable client handoff. AI performed the document processing and preparation. The client remained responsible for review and submission; the bank remained responsible for confirming its records and deciding the next action.
Result
The completed output was a transaction-specific inquiry package with preserved evidence and explicit unresolved questions. The reviewed record establishes preparation. It does not establish that this package was sent, accepted by the bank, or followed by credit to the beneficiary.
There is no claim here of funds released, revenue generated or a deployed autonomous clearance system.
Transferable lesson
Incomplete evidence can still support a precise next action. The useful design decision was to define what each document could establish, keep the unresolved relationships visible and address those questions to the party able to answer them.
For founders using AI across complex business work, that is a practical operating principle: build the handoff around the next decision, its evidence and its owner. A polished explanation is only useful if it preserves what is still unknown.