Beyond Impersonation: Proactive Supplier Vetting to Stop Advanced Business Email Compromise
The invoice lands at 4:47pm on a Friday. Supplier name correct. ABN correct. Purchase order number matches the one raised three weeks ago. Only the bank account has changed — and the attached letterhe
The invoice lands at 4:47pm on a Friday. Supplier name correct. ABN correct. Purchase order number matches the one raised three weeks ago. Only the bank account has changed — and the attached letterhead bears the new details on what looks like official stationery.
Your AP clerk processes it. The payment clears Monday. By Tuesday the real supplier calls asking why they haven't been paid. The money sits in a mule account in western Sydney, already fragmented across three crypto off-ramps. ASIC's 2023–24 enforcement outcomes show business email compromise losses — but that figure only captures what gets reported. Most organisations absorb the loss quietly, write it off as a processing error, and tighten a checklist that failed the first time.
This isn't a typo you were meant to catch. It's a supply chain compromise that began months earlier when a credential harvest gave attackers the language, timing, and document templates to make the request feel routine. The failure isn't human. It's systemic: a payment workflow that treats an email attachment as sufficient authority to move six figures.
The attacker didn't guess the invoice format — they harvested it. A compromised mailbox at the supplier's end, or a subcontractor's project portal, gave them three months of legitimate correspondence: the exact phrasing the accounts team uses, the approval chain cc list, even the internal reference codes that only appear on page two. They know your cutoff is 3pm Friday. They know the CFO authorises anything over $50k and the CFO is at a conference this week. The request arrives calibrated to every control you think you have.
Standard email gateways check sender reputation, SPF alignment, attachment sandboxing. They don't check whether the person hitting "send" is the same person who negotiated the contract six months ago. They can't verify that the PDF letterhead wasn't reconstructed from a legitimate invoice uploaded to a tender portal in 2022. The compromise lives in the gap between "this email passed technical checks" and "this instruction is genuinely authorised" — and every payment workflow that relies on the former to satisfy the latter is already exposed.
Beyond the Domain: Deconstructing Modern Email Impersonation Vectors
The attacker doesn't need to spoof the domain when they already own the mailbox. Credential compromise — phishing, credential stuffing, session hijacking — puts the threat actor inside the supplier's actual tenancy, sending from a legitimate address with a valid DKIM signature, passing every DMARC check your gateway runs. The email arrives in-thread, referencing the correct purchase order, signed with the same footer the accounts team has used for years.
Lookalike domains still work, but they're the blunt instrument. The precision play is thread hijacking: the attacker monitors a live correspondence chain for weeks, waiting for the payment instruction moment, then replies from the compromised account with "updated banking details — please action urgently." Your gateway sees a trusted sender, a clean reputation score, and a conversation history that checks out. The technical controls verify the pipe, not the person at the other end.
Gateway vendors market AI anomaly detection, but the models train on volume — bulk phishing, malware payloads, known bad IPs. A single reply in an established thread, sent during business hours from a device fingerprint that matches the supplier's usual login pattern, carries no statistical signal. The attacker doesn't need to mimic writing style when they can quote the last three emails verbatim and attach the genuine invoice PDF with only the BSB and account number altered in the footer. Your three-way match validates the PO and the goods receipt; it does not validate that the banking details on the PDF still belong to the supplier.
Session tokens bypass MFA entirely. Once the attacker exports a valid session cookie — harvested via adversary-in-the-middle phishing kits like Evilginx or purchased on Genesis Market — they authenticate directly to Exchange Online or Google Workspace without touching a password prompt. Conditional access policies evaluate the token as compliant. The email originates from the supplier's IP range, passes SPF/DKIM/DMARC alignment, and lands in the inbox with a green trust banner. The technical failure isn't in the gateway; it's in the assumption that cryptographic verification of the transport layer equals verification of the human intent behind the payment instruction change.
Compromised credentials remain the highest-leverage vector because they inherit the supplier's entire trust footprint. An attacker with valid O365 or Google Workspace credentials sends from the genuine tenant, on the genuine IP range, with genuine DKIM signatures. No gateway rule flags the message because, cryptographically, nothing is wrong. The failure point shifts to the identity provider — token replay, legacy authentication protocols still enabled on service accounts, and the absence of phishing-resistant MFA (FIDO2/WebAuthn) on finance-adjacent mailboxes. Consider a typical scenario: surveillance of email compromise losses shows that a majority of reported BEC incidents involve credential theft rather than domain spoofing, yet most AP teams still train staff to check sender addresses.
Lookalike domains persist as a parallel channel, but the technique has matured. Attackers register domains with valid SSL certificates, publish SPF/DKIM/DMARC records that pass alignment, and warm the IP reputation over weeks before the first malicious send. Some operate fully functional mail servers that handle legitimate correspondence for weeks, building sender reputation, before pivoting to a single payment diversion request. The gateway sees a "trusted" sender with clean history. The only reliable signal is behavioural: a first-time banking detail change from a supplier with whom you've transacted for years, regardless of how perfect the email container appears.
The Protocol Layer: Building Out-of-Band Verification into Payment Workflows
No email thread should ever authorise a banking detail change — not even one that passes every cryptographic check. The only control that survives credential theft, domain spoofing, and compromised mailboxes is an out-of-band confirmation to a pre-registered contact using a channel the attacker cannot intercept. That means a phone call to a number captured at onboarding, not one printed on the invoice or embedded in the signature block.
ASIC treats payment instruction changes as high-risk variations requiring documented, multi-person sign-off. A single AP clerk calling a mobile number supplied in the same compromised email chain is not a control; it is theatre. The workflow must enforce: initiator verifies against the master supplier record, a second authorised officer independently confirms the change via the onboarding contact, and both sign the variation log before the ERP record is updated. Anything less leaves a gap that no gateway can close.
Most fraudulent banking changes arrive with urgency baked in — "effective immediately" or "before month-end run" — because pressure degrades the very controls designed to catch them. A robust workflow bakes in a mandatory cooling period: no banking update takes effect within 48 hours of verification, giving the genuine supplier time to flag unauthorised requests. This single delay has stopped more fraud than any spam filter.
The variation log itself must be immutable — stored outside the email system, timestamped, and linked to the supplier master record in the ERP. When ASIC examiners request evidence of verification under the Corporations Act, a screenshot of an email thread is not a record; a signed, dated variation form with two named officers and the onboarding phone number dialled is. If your process cannot produce that document in under five minutes, the control does not exist in practice.
The phone call rule only works if the number dialled was captured at onboarding — not the one printed on the invoice, not the one in the email signature, not the one the caller offers when you hesitate. Supplier master records should carry a verified contact name, direct line, and extension, confirmed by a test call before the first payment runs. When a banking change request arrives, the authorised officer dials that exact number, asks for the named contact, and confirms the BSB and account digits verbally. No voicemail, no callback to a different number, no "I'll have accounts call you back." If the contact has left, the process escalates to the onboarding signatory, not the replacement offered in the email.
Two-person verification is not a suggestion — it is a minimum control that helps directors meet their duty under the Corporations Act to exercise care and diligence. The initiator verifies against the master supplier record, a second authorised officer independently confirms the change via the onboarding contact, and both sign the variation log before the ERP record is updated. Anything less leaves a gap that no gateway can close.
The variation log itself must be immutable — timestamped, linked to the supplier master record, and stored outside the email system where the fraud originated. A PDF in a shared drive satisfies auditors; a write-once ledger entry in the ERP with dual digital signatures satisfies directors' duties under s 180(1). The log captures: the original banking details, the requested change, the onboarding contact dialled, the verbal confirmation received, the two officers who verified, and the exact time the ERP record was updated. No free-text fields, no "notes" column that swallows the evidence.
Quarterly spot-checks by internal audit — or the external auditor — should test a sample of supplier banking changes against that log. The test is binary: does every field populate, and does the dialled number match the onboarding record? A single "no" is a control deficiency reportable to the audit committee. The protocol layer is not a checklist; it is a forensic trail built before the fraudster sends the first email.
Uncommon Insights
Most AP teams verify the invoice. Almost none verify the entity behind it. A supplier emails a bank account change — logo correct, tone familiar, history clean — and the payment goes out. The fraud isn't in the email. It's in the onboarding file that nobody reopened. The ABN matches. The trading name matches. The registered address on ASIC Connect is a serviced office in Parramatta the supplier vacated eighteen months ago. The phone number on the invoice rings through to a mobile that forwards to a call centre in Manila. None of this appears in the ERP because the master record was never refreshed after the first payment.
Picture a mid-sized construction firm in western Sydney. A long-standing subcontractor requests a banking update via email — same project manager, same reference format, same courteous sign-off. The AP officer follows the callback protocol: dials the number on file, reaches voicemail, leaves a message. The subcontractor calls back from a different number, confirms the change, cites the project code. $118,000 clears to a mule account in Brisbane before the real subcontractor queries the missing payment. The failure point wasn't the callback. It was the absence of a pre-authorisation check: an ASIC current extract would have shown the subcontractor's principal place of business changed six months prior; a credit bureau alert would have flagged a new directorship linked to a previously disqualified director. The intelligence existed. The workflow didn't require it.
ABN Lookup and ASIC Connect are free, but they are point-in-time snapshots. A supplier's status can shift between onboarding and the next payment run — voluntary administration, wind-up notices, director disqualifications, or a change of principal place of business to a residential address in a different state. Continuous monitoring bridges that gap. Credit bureaus offer supplier risk feeds that push alerts on court actions, defaults, and ASIC published notices directly into procurement platforms. The ATO's Business Registry Services API can validate ABN status and GST registration in real time, not weeks later during reconciliation.
Geospatial verification adds another layer. Cross-referencing the supplier's declared address against Landgate, NSW Land Registry Services, or equivalent state title databases confirms whether the premises is commercial, industrial, or a residential front. A subcontractor listing a factory address that title records show as a single dwelling raises an immediate flag — one no three-way match catches because the invoice, purchase order, and goods receipt all align. The $118,000 loss in western Sydney occurred because the workflow treated onboarding as a one-off event. Treating supplier intelligence as a continuous control, not a periodic audit task, shifts the detection point from post-payment recovery to pre-authorisation prevention.
Beneficial ownership opacity remains the hardest vector to close. ASIC's register of registrable controllers captures ultimate beneficial owners for proprietary companies, but the data is self-reported and rarely refreshed after incorporation. A supplier's director may hold shares through a chain of three discretionary trusts — none of which appear on the ASIC extract — masking a conflict of interest or a sanctioned entity. The ATO's Client Identity Information register and AUSTRAC's beneficial ownership guidance for obliged entities both expect deeper verification, yet most AP teams stop at the ABN lookup. Running the supplier's ACN through the ASIC Connect "Organisation and Business Names" search reveals hidden links that a single director search misses.
Insolvency intelligence is the final pre-payment gate. ASIC and AFSA insolvency registers update daily, but few ERP systems ingest them automatically. A supplier entering voluntary administration on Tuesday can still receive a progress payment on Thursday if the AP team relies on month-end supplier master file reviews. Embedding an API call to the insolvency notices endpoint at the moment a payment batch is generated — rejecting any supplier with an active external administrator appointment — costs pennies per transaction and eliminates the "we didn't know" defence. The western Sydney loss would have been blocked at that gate: the phishing entity had an active winding-up application lodged six weeks before the fraudulent invoice was submitted.
Physical address verification closes the last gap. A supplier listing a residential address in a quiet suburb while invoicing for commercial construction volumes should trigger an immediate hold. Cross-referencing the ABR-registered principal place of business against publicly available property and mapping resources takes minutes and catches the shell entities that ASIC company searches alone miss.
The western Sydney subcontractor loss traces directly to skipping these gates. The phishing entity operated from a residential rental in Blacktown, had an active winding-up application (ASIC Published Notice 505 lodged 42 days prior), and listed a director disqualified under the Corporations Act 2001. Three independent public registers — ASIC Insolvency Notices, ASIC Disqualified Persons Register, NSW LRS title data — each would have blocked the payment. None were checked. The $118,000 didn't vanish because of a clever email; it vanished because the AP workflow treated onboarding as a paperwork exercise instead of an intelligence function.
Compliance Checklist: Hardening the AP Function Against Fraud
DMARC enforcement set to "reject" is the baseline — not "quarantine", not "none". If your domain publishes a DMARC record without a reject policy, you are advertising that spoofed emails from your own suppliers will reach your AP inbox. Pair that with SPF hard fail (-all) and DKIM alignment on every authorised sending service, including third‑party invoicing platforms. Most mid‑market ERPs still send from shared IPs that fail alignment; fix the sending infrastructure or isolate those flows in a subdomain with its own enforced policy.
Three‑Way Match must extend beyond PO, receipt, and invoice. The fourth leg is sender identity: the email address, domain authentication result, and the verified contact record in your vendor master file must all align before the invoice touches the payment run. A single mismatch — domain passes DMARC but the sender isn't the authorised billing contact — triggers an automatic hold and routes to the documented approval workflow for bank‑detail changes, which requires two authorised signatories and an out‑of‑band callback to the supplier's registered phone number.
Approval workflows for bank‑detail changes need a documented chain that survives auditor scrutiny. Every request — whether it arrives by email, portal, or phone — must be logged with timestamp, requestor identity, verification method, and the two signatories who authorised the update. The callback uses the supplier's registered landline from onboarding, not a mobile number supplied in the change request itself. That single rule defeats the classic "new phone number, new bank account" combo that bypasses weaker controls.
Segregate duties so the AP clerk who processes the invoice cannot also approve the master‑file amendment. In practice, that means the ERP role that edits vendor bank details is assigned to a treasury or finance lead, while AP retains read‑only access. Any emergency override requires a third signatory — typically the CFO or delegate — and a written justification retained for seven years under s286 of the Corporations Act.
| Control | Requirement | Evidence Retained |
|---|---|---|
| DMARC policy | p=reject on all sending domains | DNS record screenshot, aggregate reports |
| SPF/DKIM alignment | Hard fail (-all); DKIM aligned on all services | Validator output, third‑party sender attestation |
| Four‑way match | PO, receipt, invoice, sender identity | Automated hold log, mismatch alerts |
| Out‑of‑band callback | Registered landline only; two signatories | Call recording or signed verification form |
| Segregation of duties | AP processes; treasury approves master‑file edits | ERP role matrix, access review logs |
| Emergency override | CFO + one other; written justification | Signed memo, seven‑year retention |
Quarterly red-team exercises expose gaps that static controls miss. Simulate a compromised supplier account sending a bank-change request from a valid domain with aligned SPF/DKIM — if your AP team processes it without a callback, the control set is theoretical. Log every simulation result against the control matrix above; a 100% pass rate means the test was too easy.
Embed the verification form directly into the ERP workflow so the "call recording or signed verification form" cell in the table isn't a manual afterthought. When treasury approves a master-file edit, the system should require the callback reference number before the save button activates. That reference ties to a recorded line or a wet-ink form scanned and attached — evidence that survives an ASIC inspection under s1308 without anyone hunting through inboxes.
The checklist is useless if it lives in a PDF on a shared drive. Each row must map to a system-enforced gate: DMARC rejection at the gateway, three-way match logic that blocks payment when the sender domain fails alignment, and a master-file change workflow that physically prevents save without the callback reference number attached. ASIC expects controls to be "designed and operating effectively" — design is the ERP configuration; operating is the quarterly red-team proof.

Treat the table as a living control matrix, not a reference document. Assign each control a control owner, a test frequency, and a failure escalation path. When the red-team simulation fails — and it will — the gap feeds straight back into the matrix as a remediation ticket with a due date and a re-test trigger. That closed loop is what separates a compliance artefact from a control environment that survives a s1308 production notice.
Key Takeaways
Compliance isn't a document. It's a process enforced at onboarding — the moment a supplier enters your master file, not the moment an auditor asks to see it. Every control in the matrix above fails if the first payment instruction change bypasses the callback gate because "we know this vendor."
The $118,000 loss didn't happen at payment. It happened three months earlier when the bank details changed on a Friday afternoon and the AP clerk skipped the verification call to clear the queue.
Reactive auditing finds the hole after the money moves. Proactive supplier intelligence stops the change request before it reaches the payment queue. That means continuous monitoring of ABR status, ASIC deregistration flags, and court judgment feeds — automated into the ERP so a supplier's risk score updates the same day a winding-up application is filed, not six months later at renewal.
The callback reference number is the audit trail that matters. Every bank detail change carries a unique verification ID: who called, what number they dialled, the exact time, and the name of the person who confirmed. No reference number, no payment release — hard-coded in the workflow, not a manager's discretion. Regulators expect contemporaneous records, not retrospective explanations.
The onboarding questionnaire that sits in a shared drive, untouched since 2019, is not a control. It's a liability. Real onboarding means the supplier's ABN, registered address, and director identities are verified against live ASIC and ABR feeds before the first purchase order is cut — and that verification refreshes automatically every 90 days. If a director resigns or a winding-up notice appears, the ERP flags the vendor record and blocks new payments until a human reviews the change.
Three-way matching that only reconciles PO, receipt, and invoice amount misses the fraud vector entirely. The match must extend to identity: the bank account on the invoice must resolve to the legal entity on the ASIC extract, confirmed by an out-of-band callback logged with a reference number. Anything less is theatre. The CFO who signs off on a payments process without that identity link owns the next loss personally — ASIC's s1308 notices and APRA's CPS 234 both treat the absence of contemporaneous verification records as a governance failure, not an operational oversight.
Compliance isn't a document. It's a process enforced at onboarding, refreshed automatically, and verified out-of-band every time money moves. The organisations that stop BEC don't have better firewalls — they have better supplier intelligence.
- Onboarding is the only control point that matters — verification against live ASIC and ABR feeds before the first PO, with automated 90-day refreshes, prevents compromised vendors from ever entering the payment stream.
- Three-way matching must include identity — reconciling PO, receipt, and invoice amount without confirming the bank account resolves to the legal entity on the ASIC extract is theatre, not control.
- Out-of-band callbacks need audit trails — every payment instruction change requires a documented phone confirmation logged with a reference number; undocumented verbal approvals don't exist in a regulator's eyes.
- Reactive auditing finds losses; proactive intelligence prevents them — monitoring for director resignations, winding-up notices, and address changes in real time shifts the function from cleanup to prevention.
- The CFO owns the next loss personally — ASIC's s1308 and APRA's CPS 234 treat absent contemporaneous verification records as governance failures, not operational oversights.
Run a free supplier check in seconds
Search by business name, ABN, or ACN. Instant PASS/WARN/FAIL across 8 verification signals.
Start verifying →


