On 30 July, Japan’s National Police Agency reported that losses to phone and online scams reached ¥181.6 billion in the first half of 2026, ¥61.9 billion more than a year earlier and the worst half-year on record. That is roughly ¥1 billion a day. Investment scams run through social media led with ¥79.8 billion; calls from fake police officers took another ¥50.8 billion (National Police Agency, provisional figures; Jiji Press, 30 July 2026).
The response is getting smarter. This autumn, Macnica and Elith reported what they call Japan’s first proof of concept for spotting fraudulent calls with AI that runs on the home gateway a landline plugs into, analysing the call as it happens rather than in the cloud. In their test environment it caught more than 90% of the scam calls (Macnica).
That is real progress, and it is aimed at the right moment: the seconds before someone moves money because a convincing voice told them to. Trade has the same moment, and it mostly still relies on people noticing.
Every scam asks the same thing: believe me, and move the money now.
The fake police officer, the grandson in trouble, the investment adviser on a messaging app: the stories change, the request does not. A voice with authority asks for money to go somewhere new, quickly, before anyone checks.
Detection tries to answer one question: is this caller real? It is a good question. It is also one that the scammer gets to retry, with a better script, every day.
Trade’s version arrives by email.
In trade, the scam rarely sounds like a stranger. It reads like your supplier. An email from a familiar address, or one letter off it, says the company has changed banks and the next payment should go to a new account. The invoice is real, the goods are real, and the payment leaves on time, to the wrong place.
The FBI’s Internet Crime Complaint Center put reported business email compromise losses at about $3 billion in 2025, second only to investment fraud (FBI IC3 2025 Internet Crime Report; SpyCloud summary). Reported losses are the floor, not the total.
The instruction travels separately from the goods
Bank details sit in an email, a PDF or a spreadsheet, apart from the bill of lading and the shipment. Change the email and nothing about the cargo looks wrong.
The deadline does the persuading
Trade payments run on cut-offs: the vessel is loading, the documents are due, the bank closes at three. A request that arrives under that pressure gets less checking, not more.
Better detection helps. It still leaves a person to be fooled.
Call-back procedures, verified contact lists and AI that flags an unusual request all reduce the damage, and every exporter should have them. But each of them ends the same way: a person decides whether the instruction is genuine. The scammer only has to win that decision once.
The stronger fix is to make the decision unnecessary: a payment that does not depend on believing anyone at the moment it is released.
Fix the payee when the deal is made, not when the money moves.
If the account that will receive the money is agreed and recorded when the contract is set up, a later email cannot change it. Any change becomes a visible amendment that both sides approve, not a quiet edit in one inbox.
Release on what happened, not on what was said
If the money is released when the shipment reaches an agreed point, such as loading or the transfer of the electronic bill of lading, then the trigger is a recorded event, not a message. Nobody can phone in a new trigger.
Hold the money where both sides can see it
When the buyer’s payment is held against the deal rather than sent ahead on trust, a fraudulent instruction has nothing to redirect. The money waits for the goods and goes to the party the deal already names.
The payment should check the shipment, not the sender.
Phone-scam detection asks whether the caller is real. Trade can ask a simpler question, one that a scammer cannot answer from a distance: did the goods actually move? A message can be faked in minutes. A recorded loading, a transferred electronic bill of lading and a payee fixed at contract time are far harder to fake together.
That does not make fraud impossible. It moves the attack from one convincing email to the shipment records themselves, which is a much harder target.
This is the layer we are building.
At STANDAGE, Digitrad Payments holds the buyer’s payment against the deal and releases it on recorded shipment events, with the transfer of the electronic bill of lading as the anchor, to the seller named when the deal was set up. The seller is paid when the goods and the title have moved; the buyer’s money cannot be redirected by a well-timed email in between.
The litmus test
Ask one question inside your own company: how many people can change a supplier’s bank details, and could they do it on the strength of a single email? If the answer is “several” and “yes”, your payments are protected by attention, not by design.
For exporters, forwarders and trade banks: when a payment goes to the wrong account, who should carry the loss, the buyer who sent it, the seller who never received it, or the process that allowed one email to decide?
