Deepfake Scams and Small Business: What to Do When You Can't Trust the Voice on the Phone

The phone call is the part that gets people. An email asking you to change a supplier's bank details is the kind of thing most small business owners have learned to be suspicious of. So you do what you were told to do: you ring the person to confirm. They answer, they sound exactly like themselves, they confirm the change. You pay. Three weeks later the real supplier asks where their money went.

That second step, the phone call, was the safety net under almost every payment fraud warning issued in the last decade. In 2026 it is the part attackers have learned to fake.

Pindrop's 2026 Deepfake Readiness Index, released on 28 September 2026, surveyed more than 250 security leaders at US companies. 74% had encountered a suspected deepfake attack in the past year. Only around 10% were using deepfake detection tools. One in four of the organisations that were hit lost USD $1 million or more to a single incident, and 37% said one successful attack could put a company like theirs out of business.

Those numbers come from large organisations with security teams. That is worth sitting with for a second, because it means the businesses with the most resources are the ones reporting they are not ready. If you run a business with eight staff and no IT person, the question is not whether you have a deepfake defence. You do not. The question is whether the process you use to approve money is good enough to survive a convincing imitation of someone you know.

This is a guide to fixing that, and it does not involve buying detection software.

What Actually Changed

Voice cloning is not new. What changed is the cost and the effort.

McAfee's research team demonstrated back in 2023 that three seconds of audio was enough to produce a usable clone of someone's voice, and that one in four of the 7,054 adults it surveyed had either experienced an AI voice scam or knew someone who had. In 2023 that was a research finding. In 2026 it is a product feature on tools anyone can sign up for in an afternoon.

Three seconds is the detail that matters for a small business. It is not a technical barrier, it is nothing at all. Three seconds of audio is a voicemail greeting. It is the "hi, you've reached" on your mobile. It is a 15 second clip of you talking about your business on a local news segment, a podcast appearance, an Instagram reel, a webinar recording, the video on your own About page. For a tradesperson, agency owner or practice principal who has spent years building a public presence, the training data is already published and it is not coming back down.

In Australia the losses are already showing up. AI firm AIBUILD estimated that voice cloning scams cost Australians around $25.8 million in the first half of 2025 alone. That is an industry estimate rather than an official figure, but it is pointing in one direction. And voice cloning lands on top of an attack that already works. Payment redirection usually starts with a real email account that has been taken over, often with a password that leaked somewhere else first. The standard advice after that kind of scam has always been to ring and confirm. Voice cloning goes after exactly that step.

So the chain looks like this now. The email looks real because it came from a real compromised account, often one whose password was already sitting in a stealer log. The voice sounds real because it was cloned from material you published yourself. And the request looks real because it arrives at a moment when that request would be normal.

Why Smaller Businesses Are the Softer Target

There is a reasonable assumption that fraud like this gets aimed at companies with more money. The economics do not really work that way.

A business with 400 staff has a finance team, a procurement process, a dual approval rule for payments over a threshold and a vendor master file that somebody owns. None of that is clever technology. It is just separation of duties, and it exists because the business got big enough that nobody could be trusted to do all of it alone.

A business with eight staff has none of it, for perfectly good reasons. The owner approves payments. The owner also answers the phone, quotes the jobs and does the BAS. There is no second approver because there is no second person whose job that is. Speed is the point of being small, and a request from a known supplier on a Thursday afternoon gets handled on Thursday afternoon.

The Australian scam data reflects the pattern. The National Anti-Scam Centre's Targeting Scams report for 2024 found Australian businesses lost $19.9 million to scams that year, and around two thirds of that came from small businesses. Those are reported losses only, from businesses that knew to report. The real figure is higher, because a lot of small business fraud gets quietly written off as a bad month.

And one more thing works against smaller businesses. At your size, the person asking is not an abstract name in an org chart. It is someone you know. You recognise the voice because you genuinely do know what they sound like. That familiarity is the thing being weaponised.

The Three Requests That Matter

Deepfake fraud against a small business is almost never exotic. It comes down to three asks.

Change where the money goes. A supplier, contractor or employee emails new bank details, often with a plausible reason attached: new accountant, switched banks, old account frozen. The follow-up call confirms it. This is the single most common and most expensive version, and the one cyber insurance often covers separately, with its own limit, or not at all.

Approve something urgent and slightly out of process. A call from the owner, the bookkeeper or a senior staff member while they are travelling, asking for a payment to be made today, with an explanation for why the usual steps need skipping. Urgency plus a known voice plus a reason not to double check.

Give someone access. A call to whoever handles the computers, asking for a password reset, a new MFA device, or to be let back into an account. This is the one that turns a single fraud attempt into a long stay inside your systems, the same move behind the fake help desk calls around Microsoft's passkey rollout, where the attacker's real goal is to register their own sign-in method.

Notice what all three have in common. None of them are technical attacks. They are requests made to a human being who has the authority to say yes. No amount of security software intercepts a phone call where someone with your bookkeeper's voice asks a reasonable question.

What Stopped Working

Be honest about the controls you are probably relying on, because most of them have quietly expired.

"I'd recognise the voice." This was the main one, and it is gone. Not weakened. Gone. Treat voice recognition as evidence of nothing.

"I'd ring them to check." Still useful, but only if you ring a number you already had. If you call the number in the email, or call back a number that just rang you, you are letting the other side choose the channel. That is the whole trick.

"We'd notice something was off." Sometimes. But these requests are built to sit inside normal. A supplier changing banks is normal. An owner travelling is normal. The attack is designed to be boring.

"We're too small to be worth it." A $14,000 redirected invoice is a good day's work for someone running this at scale across hundreds of businesses. You are not being singled out, which is exactly why you are in the list.

A Verification Process You Can Set Up This Week

The fix is not detection technology. It is a small number of rules that do not care whether the voice was real, because they do not treat the voice as proof. Here is the whole thing.

Any change to payment details gets verified on a number you already had. Not the number in the email. Not a number read out on the call. The number in your own records, from before the request arrived. If you do not have one on file, that is the first gap to close: get a direct number for every supplier you pay, and write it down somewhere other than your inbox.

Nominate one channel for approving money, and never change it under pressure. Whatever it is, phone to a saved number, a message in a chat you both already use, or in person. The rule is that the request and the confirmation never travel on the same channel. An email request gets a phone confirmation. A phone request gets a written confirmation. Urgency is not a reason to skip the second channel, it is a reason to be more careful about it.

Agree a challenge question with the two or three people who can authorise payments. Something you both know that has never been online or in writing. Which cafe you went to after the Thursday meeting. It sounds low tech because it is, and that is precisely why cloning a voice does not get past it.

Put a value above which two people must agree. Pick a number that actually reflects your business, whether that is $1,000 or $10,000. If you genuinely only have one person, the second approver can be your accountant or bookkeeper. The point is that no single yes moves significant money.

Tell your team the rule out loud, once. Specifically: nobody in this business will ever ask you to skip the verification step, and if someone does, including me, the answer is no and you will not be in trouble for it. Most of these attacks succeed because a staff member did not feel able to slow down a request from the boss.

Say it to your suppliers too. A short line on your invoices and in your email signature works: we never change bank details by email, and we will always confirm a change by phone on a number already on file. It protects both directions, because someone is going to try impersonating you to your clients.

That is six things, and none of them cost anything. They work because they replace the question "does this sound like them" with "did this arrive the way we agreed."

The Part That Trips People Up

Here is where this falls over in practice, and it has nothing to do with deepfakes.

Every rule above assumes you can answer a few basic questions quickly. Who can authorise a payment? Who has admin access to the accounting system, the email tenant, the domain? Which staff members are still active and which left in March but still have an account? Which supplier logins exist, and whose name is on them? When someone calls your bookkeeper asking for an MFA reset, is there a written answer to the question of who is actually allowed to request that?

Most businesses under twenty staff cannot answer those questions without a conversation and some guessing. The approval rule exists in the owner's head. Access grew by accident. There is no list. So the verification process gets written down in a document, holds for about six weeks, then drifts, because the facts underneath it were never recorded anywhere that stays current.

Vera will not decide who is allowed to approve a payment. That rule is yours, and it should live with your accountant or bookkeeper. What Vera does is keep the facts underneath that rule current, so the rule still holds in six months:

• Your team directory. Connect Microsoft 365 or Google Workspace and Vera pulls in your users, so "who is still active" is something you look up, not something you remember.
• Your devices and software. Hardware, software subscriptions and licences in one place, including whose name is on them.
• Breach monitoring. Vera checks your team's email addresses against Have I Been Pwned every day, so you hear when a credential turns up in a breach before someone uses it to send that convincing email.

Not an MSP contract, and not a document you have to remember to update. Just a current picture of your own IT, which is what turns a verification rule from something you wrote down once into something you can actually check. If that is the gap in your business, take a look at Vera at verait.io.

The Bottom Line

Deepfake scams reached small business faster than the advice did. The standard guidance, ring them to check, was written for a world where a voice was proof of identity, and that world ended quietly sometime in the last two years.

You do not need detection software and you do not need an IT department. You need a process that never relies on recognising anyone, a second channel for anything involving money, and a clear enough picture of your own business to know who is allowed to say yes.

Set it up this week. It takes an hour. The alternative is finding out your process was inadequate at the same time you find out where your money went.