Short · Fraud and phishing · 18 min read

How Do You Verify a Supplier's Bank-Account Change?

Hold the destination change until an authorized supplier representative confirms it through a contact route established independently of the request, the proposed beneficiary has been checked, and the approved details are bound to the payment released. A familiar email thread, recognizable voice or successful account-name check can contribute evidence. None establishes that entire chain. If your independent contact cannot be trusted or reached, the change remains unresolved; urgency doesn't supply the missing evidence.

The practical question is harder than “did someone call the supplier?” You need to know where the number came from, who answered, what they had authority to approve, which account they confirmed and whether that same account reached the bank. I would design the process around those questions before adding another approval box. A perfectly completed checklist can still approve the wrong thing.

This article proposes a verification and release model for an established supplier changing its receiving account. It covers the boundary between procurement, accounts payable, treasury and security. It doesn't determine who legally owns disputed money, whether a particular payment can be recalled, or which reimbursement rules apply. The examples and acceptance exercises are analytical proposals, not reported deployments or measured fraud reductions.

Start by separating the claims hidden inside “verified”

An invoice can describe a real debt while directing payment to an unauthorized account. A message can come from the supplier's real mailbox while expressing an attacker's instructions. The FBI describes criminals gaining access to legitimate billing correspondence and advises independent verification of changes to account numbers or payment procedures (Federal Bureau of Investigation [FBI], n.d., “How BEC scams work” and “Protect yourself”). That makes message plausibility a poor substitute for a payment decision.

I would separate five claims: the commercial obligation exists; the requester represents the supplier; that representative may authorize this change; the receiving account is appropriate for that obligation; and the execution uses the approved destination. Evidence for one claim shouldn't silently satisfy the others. A purchase order supports an obligation. A bank-name response supports a narrower relationship between submitted details and banking records. Neither necessarily establishes the requester's mandate.

This separation changes the work assigned to each team. Procurement can help establish the supplier relationship and authorized contacts. Accounts payable can reconcile invoices and maintain the controlled supplier record. Treasury can inspect the executable instruction and banking response. Security can investigate suspected compromise. Those are proposed responsibilities to adapt to the organization, not an assumption that every business has four separate departments.

Consider the apparently reassuring statement, “The supplier knows the invoice amount and our last payment date.” If those details were in the compromised correspondence, the statement adds little independent assurance. That is an inference from the threat model, not a claim that every attacker has read every invoice. The reviewer should ask whether a fact distinguishes the legitimate representative from someone with access to the challenged channel.

The callback is only as independent as its contact record

A callback to the number in the change request simply gives the request another opportunity to confirm itself. The USAID Office of Inspector General's business-email-compromise alert specifically advises verifying bank changes by telephone without using numbers supplied in unsolicited messages (U.S. Agency for International Development, Office of Inspector General [USAID OIG], 2023, “Mitigation Efforts”). The useful principle is independence of the contact route, not the sound of a telephone ringing.

For an established supplier, I would start with contact information validated during onboarding or an earlier controlled update. Record when and how it was established, who could alter it and whether the current requester also seeks to change it. A record is not trustworthy merely because it sits in an enterprise application. If the same unreviewed email can replace both the bank account and the callback number, the application preserves the dependency instead of removing it.

Keep contact changes and payment changes as separate decisions, but examine them together when they occur close to one another. A legitimate supplier may change both after a merger or staffing change. That calls for renewed establishment of authority, rather than automatic rejection or automatic acceptance. The process needs a route to procurement or another responsible relationship owner who can investigate the changed organization without relying entirely on the disputed request.

An incoming call is not equivalent to an independently initiated callback. The FTC explains that displayed caller information can be falsified and should not establish who is calling (Johnson, 2016). A matching number on the screen therefore doesn't repair a missing contact record. Nor should a voicemail saying “use my new mobile” update the record by itself. Preserve the distinction between a communication received and a route independently selected.

Voice recognition is also limited. The FTC's warning about cloned voices advises contacting the purported person through a known number rather than trusting the voice alone (Puig, 2023). That consumer example supports caution about the signal; it doesn't measure the prevalence of supplier voice fraud. My business-process inference is that a familiar voice can assist a conversation, but must not become the sole credential for redirecting organizational funds.

Establish authority, then confirm the complete instruction

Reaching a real employee at a real supplier still leaves an authority question. A delivery coordinator may know the order yet lack responsibility for treasury instructions. An accounts-receivable contact may legitimately direct payment to a collection arrangement that differs from the trading name. I would ask the relationship owner to establish which role may request a destination change and what evidence is required for unusual arrangements. Avoid turning an analyst's guess about job titles into policy.

The verification conversation should identify the supplier entity, the reason for the change, the effective date, the invoices or payment stream affected and the full destination appropriate to that payment route. Do not rely only on a few ending digits when several stored destinations could share them. Conversely, avoid broadcasting complete financial details in ordinary tickets. An authorized verifier needs the complete instruction in a protected record; routine observers may need only a reference to it.

The confirmation should be recorded as an outcome with provenance: who initiated the contact, how the route was selected, whom they reached, what authority was established and which instruction version was confirmed. That is more useful than “phone verified.” Recording a conversation is not inherently necessary. A structured record may suffice for the organization's acceptance needs; any recording practice needs its own privacy, retention and jurisdictional assessment.

If the representative disputes the request, stop the change and preserve both accounts of events. If the representative cannot confirm the complete instruction, classify it as unresolved. Don't coach them toward agreement by revealing every expected answer and then treating repetition as independent knowledge. The objective is an attributable authorization through an appropriate route, not a performance in which two people say matching numbers.

For a supplier whose communications are believed compromised, the ordinary callback may itself need reassessment. The next task becomes establishing a usable relationship boundary during the incident. I discuss that wider problem in responding when a supplier suffers ransomware. A bank-change ticket should not quietly decide that the supplier's entire communication environment is trustworthy again.

Use account-name checks for the question they answer

Account-name services can identify discrepancies that a manual process misses, but their outputs require interpretation. Pay.UK describes Confirmation of Payee outcomes including a match, close match, mismatch and inability to check. Its guidance also makes clear that the service does not guarantee fraud detection or reimbursement (Pay.UK, n.d., “How does Confirmation of Payee work?” and reimbursement FAQ). Keep the actual response and its context, rather than translating every non-error result into “supplier verified.”

In the SEPA context, the European Payments Council describes Verification of Payee as checking supplied account and payee data through participating payment-service providers. It explicitly distinguishes that messaging service from identifying a private or legal person (European Payments Council [EPC], n.d., “How does the VOP scheme work?”). These are different schemes with their own implementation boundaries, not a universal capability available for every country, bank, payment route or customer.

A name match doesn't answer whether your supplier authorized this invoice to be paid there. A close match is not permission to substitute the returned name and continue automatically. A mismatch might reveal an error, an attack or an arrangement requiring explanation. An unavailable result is missing evidence, not a negative fraud finding. These are proposed decision interpretations of the service limits, not claims that the schemes impose the same corporate workflow.

Where the intended beneficiary is an affiliate, factor or collection agent, resolve the relationship to the obligation. A plausible explanation in the original email cannot be the only support. Route the arrangement to the people able to validate the underlying commercial authority and obtain the appropriate documentation. I would retain the reason for accepting the difference, the evidence used and the exact destination approved, rather than creating a permanent “ignore name warnings” exception for the supplier.

The same discipline applies to a small test payment. At most, its observed receipt can add evidence about delivery to a destination and the recipient's ability to acknowledge it through the chosen route. It does not independently prove the recipient's right to receive the full debt. If the acknowledgment arrives through the suspect mailbox, you may have tested the attacker's ability to reply. I would not use a token transfer as a shortcut around unresolved authority.

Authentication protects entry; transaction authorization protects the decision

Strong authentication remains necessary for the systems that hold contacts, supplier records and payment instructions. CISA recommends phishing-resistant MFA and explains that common MFA methods have different susceptibilities to phishing and related attacks (Cybersecurity and Infrastructure Security Agency [CISA], 2022, “Overview” and “Recommended Implementations”). The useful implementation question is which administrators, finance users and service accounts can alter the critical records, and how their access is protected.

NIST defines phishing resistance in terms of preventing authenticator secrets or valid outputs from being disclosed to an impostor verifier without depending on user vigilance. It distinguishes cryptographic binding from manually entered outputs that an impostor can relay (Temoshok et al., 2025, §3.2.5). That protects an authentication event. It doesn't establish the commercial truth of a supplier instruction shown to a legitimately authenticated employee.

I would therefore test two different failures. First, can an unauthorized person access or change the relevant record? Second, can an authorized person be induced to approve a destination without adequate evidence? A control can succeed at the first and fail at the second. Treating a successful MFA prompt as proof that the payment is legitimate confuses access assurance with the meaning of the action.

The FFIEC's financial-institution guidance describes layered security and risk assessment that considers transaction characteristics, including value, volume and irrevocability (Federal Financial Institutions Examination Council [FFIEC], 2021, §§3–4). That guidance addresses financial institutions; applying its risk distinctions to a corporate supplier workflow is my synthesis, not a statement that the same supervisory obligations apply to every purchaser. It supports examining the consequential action rather than assuming all authenticated sessions are equally safe.

Bind approval to a version of the destination

The decisive technical boundary sits between verification and execution. Suppose a reviewer confirms account A, but the supplier master changes to account B before the payment file is generated. An approval attached only to the supplier's name may follow the record into a different transaction. I would require the workflow to preserve the approved destination version and make subsequent changes visible before release.

OWASP's transaction-authorization guidance calls for users to acknowledge significant transaction data, for authorization to be enforced server-side, and for execution to check the authorization. It also addresses changes to transaction data during authorization (OWASP Foundation, n.d., §§1.1, 2.1, 2.6–2.8). Those principles provide a foundation. The supplier-specific version model below is a proposed application, not a feature claimed for any particular accounting package.

The change record should identify the prior and proposed destination, supplier entity, scope of invoices or payment stream, effective date and evidence reference. Approval applies to that recorded version. A material alteration produces a new version and reopens the affected checks. Editing the bank destination, changing the beneficiary entity or widening the covered payment stream should not leave the old approval looking current merely because the ticket number remains unchanged.

At payment preparation, compare the destination used by the payment instruction with the approved version, not only with whichever master record is current at that moment. If the system cannot retain this relationship, the control owner needs an explicit compensating review of the executable instruction. That review must happen before release and be evidenced. A later reconciliation can detect the consequence but cannot retroactively authorize the instruction that left.

The following state model is an analytical design proposal. It separates a valid request from an executable payment without assigning universal waiting periods or monetary thresholds.

StateWhat it meansEvidence needed to advance
ReceivedA change was requested; no trust decision yetOriginal request, supplier reference and proposed scope
Contact establishedAn appropriate route and representative were establishedContact provenance and authority assessment
Instruction verifiedA specific version of the destination was confirmedComplete protected instruction and verification record
Change approvedInternal authority accepted that versionIndependent approval and resolved exceptions
AppliedThe controlled supplier record reflects the approved changeRecorded version and update result
Release eligibleThis payment uses an authorized destinationComparison with the executable instruction and payment approval
Closed or heldOutcome and unresolved issues are explicitBanking response, reconciliation or named hold reason

The distinction between “applied” and “release eligible” matters when files, integrations or banking templates sit downstream of the supplier master. The diagram in a procedure may show one application, while the actual payment takes several copies. Map those copies and transformations. An approval on one screen should not be assumed to protect an exported file, a manually keyed instruction or an old beneficiary template.

Separate duties without duplicating the same uncertainty

NIST's AC-5 control addresses identifying duties requiring separation and supporting that separation with access authorizations. Its discussion recognizes that the boundary can span systems and that collusion remains a limit (Joint Task Force, 2020, AC-5). For a supplier change, I would distinguish editing contacts, verifying the request, modifying the destination and releasing a payment. The exact staffing model can vary; the incompatible capabilities still need to be understood.

Two approvers reading the same email do not create two independent sources. Their value depends on what each checks and what each can change. One reviewer might establish supplier authority; another might reconcile the approved instruction with the payment. If both merely confirm that the invoice looks familiar, the second signature increases the appearance of assurance without resolving the missing claim.

Administrative access belongs in this analysis. A person who cannot approve payments through the normal interface may still alter a workflow, reset another user's access or import a beneficiary file. Conversely, a verifier may be able to update the supposedly independent phone directory. I would review those cross-system combinations with the application and identity owners. The purpose is to expose bypasses, not to claim that a particular combination is necessarily exploited.

Small organizations can face genuine staffing constraints. A workable design might use an available director or an external finance service for a specific independent check, with restricted access and clear responsibility. That arrangement needs assessment of its own dependencies. The label “external” does not make evidence independent if everyone relies on the same unverified request. Where meaningful separation is unavailable, record the remaining exposure and the bounded compensating action instead of pretending a second username solves it.

Make urgency an escalation condition, not an alternative credential

A critical supplier may threaten to pause deliveries while a change is unresolved. The cost of delay can be real. I would give the finance decision-maker a concise statement of what is established, what is missing, which payment is affected and what alternatives have been examined. “Cyber said no” is not enough information for a consequential business decision.

The escalation should distinguish a verified destination from an accepted uncertainty. An executive may have authority to accept some business exposure, but their signature cannot change an unverified bank instruction into a verified one. Preserve that distinction in the case and downstream reporting. Otherwise, exception handling teaches the system to erase the very uncertainty that management needs to see.

Possible alternatives include delaying only the disputed payment, using a previously established route after checking its continued appropriateness, or arranging another validated commercial solution. None should be automatic. The old account might be closed, restricted or no longer authorized. Splitting a payment into smaller amounts doesn't resolve destination legitimacy either. Decide within the actual contract, treasury arrangements and organizational authority; don't invent a universal safe amount or cooling-off period.

Exceptions should identify their scope and end condition. An authorization for one urgent invoice should not silently become a permanent new supplier destination. If an exception permits release, the record must state who authorized it and what evidence was unavailable. Later review can then examine whether the exception was appropriate and whether the ordinary process needs repair, without rewriting the history of the original decision.

A fictional change request through the entire chain

Consider fictional Brackenford, which receives a bank-change request in an existing supplier invoice thread. The invoice amount is correct, and an incoming caller uses the supplier's familiar contact name. The bank's name-check service returns a close match to a related-looking company. None of these facts, taken alone or together, resolves the authorization chain.

The accounts-payable reviewer holds the change and retrieves the previously established supplier contact. The record shows that the phone number was edited that morning by the same user processing the email. Rather than treating the new number as independent, the reviewer asks procurement to re-establish contact through the existing relationship evidence. This is the first decisive finding: the verification route itself had changed.

Procurement reaches an appropriate supplier representative through a separately established route. The supplier confirms a legitimate treasury reorganization but says the proposed destination belongs to a different collection arrangement. Brackenford obtains the relevant commercial confirmation and records which invoices it covers. The bank-name response is retained as a discrepancy that required explanation, not relabeled as an exact match.

The approved destination is then applied to the supplier master. Before release, treasury discovers that an older payment template still points elsewhere. The payment remains held until the executable instruction is reconciled with the approved version. The case closes only after the authorized payment outcome is recorded and the template discrepancy has an owner. This fictional case demonstrates two separate control questions: who authorized the change, and what instruction would actually execute. It claims no observed loss prevented or universal product behavior.

Test the process without moving money to prove a point

I would assess this workflow with synthetic supplier records and a controlled environment before considering any production exercise. Start with a request that changes both the contact and bank details. The expected outcome is a visible dependency and a hold pending independent establishment, not a completed callback to the newly supplied number. Record which role and system enforced that outcome.

Then change the destination after approval. The exercise should establish whether the old approval is invalidated or whether a release control detects the mismatch. Include the paths your organization actually uses: manual entry, import, scheduled payment files and beneficiary templates where relevant. A successful demonstration through the normal screen says little about an untested import route.

Other cases should distinguish a name mismatch from an unavailable check, an authorized affiliate from an unexplained beneficiary, and an unreachable contact from a negative confirmation. Test an emergency exception with a narrow scope and confirm that it cannot authorize unrelated future payments. These are proposed acceptance scenarios; they have not been empirically validated as a predictive fraud model or a complete test suite.

Measure the population of requested changes and their outcomes, including unresolved cases, rework, exceptions and verification elapsed time. Review whether releases used the approved version. Keep confirmed fraudulent attempts separate from errors and unexplained requests. A held payment is not automatically a prevented loss, and an absence of reported fraud is not proof that the controls worked. The evidence should support the specific outcome claimed.

If the payment has already gone, run response in parallel

Don't make the recovery effort wait for a perfect internal explanation. The FBI advises contacting the financial institution immediately when BEC occurs; the NCSC similarly directs affected businesses to contact their bank through official details and notify the responsible IT function (FBI, n.d., “Report BEC”; National Cyber Security Centre [NCSC], n.d., “What action should I take?”). These steps do not promise that funds can be recovered.

Preserve the original request, contact history, supplier-record versions, approvals and payment references for authorized responders. Establish whether other pending payments depend on the same changed destination or contact. Coordinate a scoped hold and investigation through the responsible finance and incident authorities. The operational question is whether the same uncertainty can produce another payment while the first is being investigated.

A suspected mailbox compromise also deserves a separate technical investigation. Email-domain authentication addresses an adjacent prevention boundary; it cannot settle whether a real supplier representative authorized a particular instruction. Avoid letting a clean-looking email result close the finance case, or a successful bank inquiry close the account-compromise investigation.

What this model can establish—and what it cannot

The strongest objection is that established contact records can themselves become stale or compromised. That's correct. Independence is a property to investigate, not a permanent label. A callback, portal, bank response and internal reviewer may share hidden dependencies. The model makes those dependencies inspectable, but it cannot prove that every supplier employee is honest or every upstream system uncompromised.

A second objection is that the process can slow legitimate changes. It can. The answer is to maintain usable contact and authority records, make the evidence requirement clear and provide a staffed escalation path. Removing the check whenever delay matters makes the control least reliable at the moment pressure is greatest. Equally, a procedure that cannot resolve ordinary legitimate changes needs redesign rather than congratulation for blocking everything.

The twelve sources establish threat characteristics, authentication and authorization principles, service boundaries and response guidance. They do not experimentally validate this combined supplier-change model. Regional banking services differ; this article does not establish legal duties, reimbursement eligibility or contractual discharge. Its contribution is a falsifiable chain of decisions: an assessor can identify where evidence was missing, where approval became stale and where execution diverged.

If your current process ends with “we called them,” I'd start by tracing one supplier change from the original contact record to the payment instruction. Discuss a supplier-payment control review. We can scope the verification steps, administrative bypasses and release evidence that need examination, then define a practical acceptance exercise before you invest in another approval layer.

References