What Auditors Ask About Supplier Bank Changes, and the Evidence to Keep
An email lands from a supplier you've paid fifty times. The logo is right. The invoice number follows the sequence. Only the BSB and account number are new.
An email lands from a supplier you've paid fifty times. The logo is right. The invoice number follows the sequence. Only the BSB and account number are new.
Your team updates the vendor master file. The next payment clears. Three months later, the real supplier calls asking why they haven't been paid. The money is gone — moved through three accounts in under an hour.
This isn't a cyber breach. It's a process failure that started with a change request nobody verified independently.
External scrutiny doesn't care about intent. When auditors trace a payment to a fraudulent account, they follow the change record in the vendor master file. They ask who approved it, what verification was performed, when it happened relative to the payment run, and how the evidence was captured. If the answer is "we replied to the email," the control gap is documented and the finding is material.
CFOs who treat vendor banking changes as administrative housekeeping miss the point. Every unauthorised amendment is a potential breach of record-keeping obligations and a direct hit to cash. The controls that satisfy scrutiny aren't complex — they're just specific, repeatable, and documented at the moment of change.
The Audit Lens: Why Vendor Master File Changes Trigger Scrutiny
Auditors don't sample vendor master file changes to tick a box. They sample them because every unauthorised amendment is a control failure that cascades into misstated cash balances, unreconciled payables, and potential breaches of record-keeping obligations under the Corporations Act. The workpaper request is always the same: show me the change request, the independent verification, the approver's sign-off, and the timestamp — all linked to a single transaction trail.
If your process relies on a spreadsheet log updated after month-end, the finding writes itself. Auditors will trace a changed BSB from the general ledger back to the master file, then ask who authorised it, when they verified it, and how they confirmed the request came from the actual supplier — not someone who compromised the supplier's email. The answer cannot be "the AP team handled it." It must be a named person, a dated action, and a verification method that doesn't share a channel with the request itself.
The most common gap isn't the absence of a policy — it's the gap between the policy and the evidence that it was followed. A supplier emails new bank details on Friday afternoon. The AP clerk updates the master file Monday morning. The approval email from the finance manager arrives Tuesday. By the time auditors request the trail, the sequence has been reconstructed to look compliant: request, verify, approve. But the timestamps tell a different story. The master file was changed before verification occurred. That timing discrepancy is the finding.
Auditors will test whether the verification actually happened before the change took effect in the ERP. They'll check if the phone number used for the callback matches the one on the original contract or onboarding pack — not the one in the change-request email. They'll ask for the name of the person at the supplier who confirmed the details, and whether that person is an authorised signatory per the contract. If the evidence log shows "called supplier" without a name, a number, and a time, it's not evidence. It's a placeholder.
The Corporations Act requires companies to keep financial records that correctly record and explain transactions, and to do so in a way that enables true and fair financial statements to be prepared and audited. A vendor master file change that bypasses verification isn't just a processing error — it breaks the chain of evidence that supports the integrity of every subsequent payment. Auditors will trace a sample of changed records back to the source request and forward to the first payment on the new details. If the authorisation sits only in an email thread, if the verifier can't be identified, if the timestamp shows the change went live before the callback was made, the control is treated as not operating. That finding flows straight into the management letter and, if material, into the audit opinion on internal controls over financial reporting.
The practical test is simple: can you produce, for any single change, a dated record showing who requested it, who verified it independently, who approved it, and the exact time the master file was updated — with each step performed by a different person and each step preceding the next? If the answer is no for even one instance, the process fails the scrutiny it will receive.
Evidence Threshold: Moving Beyond the Confirmation Email
A supplier emails new bank details. Your AP clerk replies to that same thread asking for confirmation. The reply comes back "Yes, correct." The details get updated. That sequence satisfies a policy checkbox — but it proves nothing. The email you replied to and the confirmation you received could both come from the same compromised mailbox, the same attacker, or a spoofed domain that passes a casual glance. Auditors know this. They will not accept a reply-to-same-thread as independent verification because it fails the fundamental test: you have not established that the person authorising the change is the entity you contracted with.
The evidentiary standard requires a verification channel that exists independently of the change request itself. That means a phone call to a number sourced from your onboarding file — not the invoice, not the email signature, not the letterhead attached to the request. It means a signed confirmation from a director or authorised signatory whose authority you verified at contract inception. It means a portal login with MFA tied to a known identity. The principle is straightforward: the method of confirmation must be incapable of being intercepted or replicated by the party requesting the change. Anything less is not evidence; it is theatre.
Consider the typical compromise: an attacker gains access to a supplier's accounts receivable mailbox, monitors legitimate invoice traffic, then inserts a change-of-bank-details request into an active thread. The domain is correct. The signature block is correct. The language matches prior correspondence. Your AP clerk replies to that thread asking for confirmation. The attacker, sitting in the same mailbox, replies "Confirmed." From the clerk's perspective, the loop is closed. From an evidentiary perspective, the loop never left the attacker's control.
This is why the verification channel must be bound to identity, not to the communication thread. A call to the number on the original supplier agreement — the one you tested at onboarding — forces the attacker to compromise a second, independent channel (the phone network, a director's mobile, a signed letterhead). A portal-based approval with MFA tied to a known user account requires the attacker to compromise both credentials and the registered device. Each additional independent factor exponentially raises the attacker's cost. The evidence log should record not just that you called, but which number you dialled, who answered, their role, and the exact words of confirmation. That specificity is what survives a deep-dive audit; a generic "verified by phone" note does not.
The evidentiary standard shifts when you move from "we have an email" to "we independently confirmed the identity of the person authorising the change." A phone call to the number on the original supplier agreement forces the attacker to compromise a second, independent channel. A portal-based approval with MFA tied to a known user account requires credential theft plus device compromise. Each additional independent factor exponentially raises the attacker's cost while creating a verifiable audit trail.
The evidence log must capture the mechanics of that independence: which number was dialled, who answered, their title or role, the exact words of confirmation, and the time. A generic "verified by phone" note survives neither a deep-dive audit nor a fraud investigation. The same discipline applies to written confirmations — a signed letterhead on company stationery, scanned and attached to the vendor master file record, carries evidentiary weight that a reply email never will. The test is simple: if the communication channel can be spoofed or intercepted without the genuine supplier's knowledge, it is not an independent verification.
Building the Audit-Proof Trail: The Evidence Log
An evidence log is a contemporaneous record created at the moment of verification — one row per change, immutable once written. If you cannot hand an auditor a log that shows who called which number, at what time, and what the supplier's authorised signatory confirmed, the control did not happen.
Start with these non-negotiable columns: Change Request ID (linked to the ticket or email thread), Supplier Legal Name and ABN, Field Changed (BSB/Account Number, Trading Name, Address), Date and Time of Verification (to the minute), Verification Method (dropdown: Verbal via Known Number, Portal MFA, Signed Letterhead, In-Person), Known Contact Details Used (the exact phone number, portal URL, or physical address), Name and Title of Person Confirming, Exact Confirmation Script (what was asked and the verbatim response), Verifying Officer Name and ID, and Attachment Reference (scan of signed letter, call recording ID, portal screenshot). No free-text notes column — every field is structured so a filter or pivot answers the auditor's next question instantly.
Attach the evidence at the point of creation, not as an afterthought. A call recording ID, a portal screenshot timestamped to the minute, or a scanned letterhead signed by the supplier's authorised signatory — each must be referenced in the Attachment Reference column and stored where the log row cannot be edited without breaking the link. If your AP system cannot lock a row once the attachment is uploaded, export the log daily to a write-once repository so the auditor sees the same record the verifier created.
Test the log quarterly: pick a sample of changes at random and confirm you can open every attachment, match the verification method to the contact details used, and trace the Change Request ID back to the original supplier email or portal request. If any row fails that test, the control gap is real and the log is evidence of the gap, not evidence of the control. Fix the process before the external auditor picks the same rows.
Lock the log to the change, not the calendar. Each row represents one supplier banking change — if a supplier updates details twice in a month, that is two rows, each with its own Change Request ID, verification method and attachment. Never collapse multiple changes into a single entry with a date range; auditors will treat that as a missing control. The log must also capture who approved the change in your ERP or AP system after verification, and the exact timestamp of that system update. If your ERP does not produce an immutable audit trail for vendor master file edits, the log itself becomes the primary evidence — so it must be tamper-evident, not a spreadsheet anyone can overwrite.
Standardise the verification method dropdown with a fixed list of approved methods — no free text. That discipline lets you filter instantly for the weakest method across the year and prove to an auditor that every change met the same bar. If a row shows a verbal method but the attachment is a call recording ID that cannot be played, the control failed. Build the log so that gap is visible in seconds, not discovered under cross-examination.
Uncommon Insights
The control that looks strongest on paper — a verbal callback to a known number — often fails because the "known number" was itself compromised months earlier. A fraudster who infiltrates a supplier's email can update the contact phone number on their email signature and website. Later, when your AP clerk calls that number to verify a bank change, they are speaking to the fraudster. The verification method passes, the log shows "Verbal — known number", and the audit trail looks clean. The failure is silent because the control never asked: when was this phone number last independently confirmed?
Map every bank change request to a specific commercial trigger. If the change request cannot be linked to a Purchase Order amendment, a signed variation, or a written notice on supplier letterhead referencing a specific contract clause, treat it as unverified until proven otherwise. The silent failure here is not the missing callback — it is the missing business reason. Auditors will test a sample of changes and ask: "Show me the commercial event that justified this." A log that records only the verification method, without the commercial trigger, cannot answer that question.
Single-source verification is a false comfort. A supplier who emails new bank details and then answers the phone at the number on that same email has not been independently verified — they have simply confirmed their own fraud. The control only works when the verification channel was established before the change request arrived: a phone number from a signed contract, a portal login credentialed at onboarding, or a physical letterhead address used for postal confirmation. If the only "known" number entered your system in the same month as the bank change, the trail is circular.
The deeper failure is temporal. Most fraudulent bank changes sit dormant for weeks — submitted, verified, approved — then activated on a payment run when attention is lowest. Build a mandatory cooling-off period into the workflow so that no bank detail change takes effect until after a second, independent verification separated by time. This forces a second independent touchpoint and exposes changes that were staged to slip through a single busy run. The audit question shifts from "Did you verify?" to "Show me the two independent verifications separated by time." That is the evidence a silent control cannot fake.

Key Takeaways
Start Monday by exporting every vendor master file change from the last 12 months and checking each one for a verification record that predates the change request. If the phone number used to confirm the new BSB was added to your ERP in the same week as the bank details, the control is circular — flag it for immediate remediation.
Replace any "reply-to-email" confirmation step with a mandatory outbound call to a verified number. Document the caller, the recipient, the exact time, and the specific details confirmed — "verified account ending in 4821" beats "confirmed banking details" every time.
Audit trails that rely on memory or goodwill collapse the moment a regulator asks for the working papers. The only control that survives scrutiny is one that leaves a timestamped, attributable record at every handoff — request received, independent verification executed, change authorised, re-confirmation completed at next payment run.
- Export and stress-test the last 12 months of vendor master file changes — any record where the verification contact was created at the same time as the new bank details is a circular control and must be remediated immediately.
- Eliminate reply-to-email confirmations — require an outbound call to a number sourced from a signed contract, onboarding portal, or pre-change letterhead, with caller, recipient, time, and exact account digits documented.
- Enforce a mandatory cooling-off period — no bank detail change takes effect until the next full payment cycle after verification, with a second independent re-confirmation at that cycle's start.
- Map every change request to a specific contract or PO milestone — if the request cannot be tied to a documented commercial event, treat it as unverified until proven otherwise.
- Build the evidence log into the workflow, not the review — fields for signatory names, verification method, dates, and confirmed account details must be captured at execution, not reconstructed at year-end.
Run a free supplier check in seconds
Search by business name, ABN, or ACN. Instant PASS/WARN/FAIL across 8 verification signals.
Start verifying →