Invoice Redirection Fraud: How to Spot It and Why Spam Filters Miss It
An email arrives from a supplier you've worked with for years. Same tone, same signature, a sensible invoice attached. The only difference is one short line: "We've changed banks, so please use the new details below." Pay it, and the money goes straight to a criminal.
This is invoice redirection fraud, one of the most common forms of business email compromise (BEC). It's worth understanding properly, because it is designed to slip past the defences most small businesses rely on.
What invoice redirection fraud looks like
The attacker's goal is simple. They want to persuade whoever pays the bills to send a genuine payment to the wrong account. They usually do this in one of three ways:
- Lookalike domains. The mail comes from a domain that is almost right. It might have an extra letter, a swapped character or a different country ending, such as .co instead of .co.uk.
- Display-name spoofing. The sender name says "Accounts – Smith Supplies", but the actual address behind it is something else entirely.
- Compromised accounts. The attacker has got into the supplier's real mailbox and replies within a genuine thread. This is the hardest to spot, because the message really does come from the right place.
Criminals use the same playbook for other scams too:
- CEO fraud: "I need you to make an urgent transfer – I'm in a meeting."
- Gift-card scams: a "manager" asks a colleague to buy gift cards and send the codes.
- Payroll diversion: an "employee" asks HR to update their bank details before payday.
Why traditional spam filters miss it
Classic filters look for things that are obviously wrong. That means known bad links, malware attachments, blocklisted servers and spammy wording.
A well-crafted BEC email often has none of these. There's no payload, just a polite request. It may come from a freshly registered domain with valid SPF and DKIM records, so authentication checks pass. And now that AI writing tools are everywhere, the old giveaway of clumsy grammar has largely disappeared.
In other words, the message is technically clean. What's wrong with it is its intent. Rule-based and Bayesian filters were never designed to judge that.
The warning signs to train your team on
Technology helps, but the people who approve payments are your last line of defence. These are the red flags worth drilling into anyone who handles invoices or payroll.
| Red flag | What to do |
|---|---|
| A change of bank details sent by email | Treat it as unverified until you confirm it by phone on a number you already hold. |
| Urgency, pressure or a request for secrecy | Slow down. Pressure is a tactic, and genuine suppliers will wait a day. |
| A sender domain that is slightly different | Compare it letter by letter with previous correspondence. Watch for doubled letters and swapped endings. |
| A Reply-To address that differs from the From address | Check before replying. Your reply may go straight to the attacker. |
| The right name but an unfamiliar address | Expand the sender details to see the real address behind the display name. |
| An unusual request from a senior colleague | Confirm it through a different channel, such as a call or a quick chat in person. |
A simple checking routine for payments
The most effective defence is a short, boring process that everyone follows every time.
- Don't act on emailed bank-detail changes alone. No new or changed account details get used until someone has verified them.
- Call back on a number you already have. Never use the phone number in the email or its signature, because the attacker may have planted it there.
- Get a second pair of eyes. Require a second person to approve any new payee, and any changed details above a threshold you set.
- Use your bank's payee name check. UK banks offer Confirmation of Payee, which checks whether the account name matches the one you entered. Treat a mismatch as a stop sign, not a formality.
- Test first. For new or changed accounts, send a small test payment and confirm receipt by phone before sending the full amount.
- Write it down. A one-page policy that everyone knows makes it easy for staff to say "I have to check" without feeling awkward.
If you think you've been caught out, contact your bank immediately. Speed matters when trying to recover funds. Then report it to Action Fraud, or to Police Scotland if you're in Scotland.
Where email technology helps
Protect your own domain
Publish SPF and DKIM records and a DMARC policy. Together, these make it harder for criminals to send mail that appears to come from your exact domain, which protects your customers and suppliers as well as you.
They won't stop lookalike domains, though. An attacker who registers a near-copy of your supplier's domain can set up perfectly valid authentication for it. That's why inbound filtering still matters.
Filter for intent, not just content
This is the gap MxGuard was built to close. It works in two stages:
- Stage one: a machine-learning classifier scores the easy 95% of mail in milliseconds.
- Stage two: messages that land in the uncertain band go to Claude, Anthropic's large language model. Claude reads each one the way a person would, looking at intent, urgency cues, impersonation patterns, lookalike-domain references and what the email is asking the reader to do.
Other checks are aimed squarely at BEC tactics:
- Lookalike domain detection covers homoglyph attacks, doubled letters and ccTLD swaps. It checks the envelope sender, the From header and the Reply-To address.
- Header analysis includes From/Reply-To divergence and display-name impersonation among the signals the classifier weighs.
- Per-domain thresholds let you set different sensitivity for different addresses. You can be strict on accounts@ and more lenient on marketing@.
Every verdict comes with a plain-English reason. Here is an example from the features page:
"The message asks the finance team to act urgently on an invoice from a domain that closely resembles a known supplier but is not a match. Pressure language plus domain similarity is a common pattern for invoice-redirection fraud."
That transparency is useful in its own right. Your team can see why something was held, which reinforces the warning signs above. Quarantined messages also appear in a daily digest email with one-click release, so a genuine invoice is never more than a click away.
No filter replaces a good checking routine, especially where a supplier's real account has been compromised. The strongest setup combines the two: filtering that catches most attempts before anyone sees them, and a payment process that stops the rest.
If you'd like the technical detail on how messages are scored, read how MxGuard works.
See how MxGuard handles your own mail
MxGuard Business costs £18.99 per domain per month, with 25 mailboxes included. It is UK-hosted, GDPR-aligned and billed month-to-month with no setup fees. Nothing changes on your existing mail server: you simply point your MX records at MxGuard.
You can try every feature free for 30 days on up to five domains, with no credit card required. Start with the MxGuard product overview, or go straight to pricing and the free trial.