BIGGER PICTURE Public & everyday services
The bank sees a scam before the money leaves
A scam victim can pass every identity check and still send money to a criminal. TSB used AI-supported network analysis to look beyond the sender and examine where a payment was going.

- Reported result
- 20% increase in APP-fraud detection
- Evidence
- Mastercard’s 2023 report on TSB
- Intervention point
- Before an account-to-account payment is sent
How the idea works
Inspect the destination network
Payment histories expose links to previously suspicious accounts.
Score before transfer
AI returns a risk signal in time for the bank to act.
Check with the customer
The bank combines its evidence with an appropriate intervention.
The customer really did authorise it
In authorised push payment fraud, the victim makes the transfer personally—perhaps believing an invoice, a supposed bank employee or a romantic partner. Authentication alone cannot establish whether that person has been deceived. The useful question becomes whether the receiving account looks connected to a wider pattern of fraud.
Sources: [2]
Follow both ends of the payment
Mastercard’s Consumer Fraud Risk combines transaction information with knowledge of accounts linked to scams. Amounts, payment history and relationships between receiving accounts help produce a risk score before a transfer proceeds. The bank uses that signal alongside its own information to decide whether to intervene. AI extends the view beyond one isolated transaction.
Sources: [1]
A concrete improvement, reported by the supplier
Mastercard’s October 2023 white paper reported a 20% increase in TSB’s detection of this fraud type. That is a meaningful operational signal, but not an independently randomized estimate. The much larger UK-wide savings figure in announcements was an extrapolation of what might happen if other banks matched TSB, not money already recovered.
Make prevention part of a human response
A later Mastercard account describes banks combining real-time intervention with customer education. That matters because a risk score is not a verdict, and legitimate payments must still work. The transferable idea is to detect suspicious relationships early enough for a useful conversation or check—before a trusted-looking request becomes an irreversible loss.
Sources: [3]