Romance fraud is not a new category. Action Fraud has tracked it for years. UK Finance includes it in annual APP reporting. But something shifted in the past eighteen months that matters for PSP fraud ops teams specifically: romance-driven APP losses have grown to a scale that now sits clearly in second place behind investment scams by total annual value lost. The £300M figure referenced in PSR submissions and Action Fraud's own reporting for the most recent annual cycle is not a headline anomaly — it reflects a structural pattern.
What that pattern means for payment interception is the focus of this piece. Romance scams have a property that most other APP fraud categories don't: a long, observable conversation window before the payment instruction arrives. That window is both the reason victims transfer large sums willingly and the reason it's the category where well-placed interception tools have the most leverage.
Why Romance Fraud Transfers Are Large
The average value per romance fraud APP transfer significantly exceeds the average for purchase scams or impersonation fraud. The reason is not that romance fraud victims are unusually trusting or vulnerable in ways that other victims aren't — it's that the scam architecture is designed to accumulate trust before requesting money, and accumulated trust translates to larger requested amounts.
A typical romance fraud engagement runs for weeks or months before the first financial request. The scammer builds a plausible identity, establishes emotional dependency, and then introduces financial need through a crisis narrative — medical emergency, business deal gone wrong, customs fees on a package containing valuables. The first transfer is often modest, partly as a proof-of-commitment test, partly to establish that the channel works. Subsequent requests escalate in line with demonstrated compliance.
By the time a fraud ops analyst at the receiving PSP looks at the transaction, it frequently shows as a single large transfer with clean account history. The victim has been preparing this payment for days. They've been coached on what to say if their bank queries it. The SCA notification arrived on their phone and they confirmed it without hesitation. From a signal perspective, this looks identical to a legitimate large personal transfer.
This is why post-payment controls — transaction monitoring, velocity checks, payee risk scoring — have limited effectiveness here. The transaction itself is the intended outcome of a carefully managed process. You're not catching an anomaly; you're catching the conclusion.
The Conversation Window and Why It Matters
Romance fraud's long conversation window is, paradoxically, an advantage for interception — provided interception is happening at the right layer.
The scammer has to maintain contact with the intended victim over an extended period. That sustained engagement has fingerprints: device identifiers associated with multiple fraud attempts, phone number metadata that doesn't match claimed location or identity, message cadence patterns that don't match genuine relationship dynamics, platform-switching requests designed to move the conversation to channels with less oversight.
A PSP that is monitoring the incoming communication context — not just the outgoing payment instruction — has signals available throughout that window. The challenge is that most PSPs don't have tooling at that layer. Their fraud stack sits at the payment rail, not upstream at the conversation. By the time the payment arrives, the conversation window has done its work.
We're not suggesting that PSPs should be reading customer communications wholesale — RIPA 2000 has specific provisions around lawful interception that govern what's permissible at the PSP's instruction, and we've written about that framework separately. The point here is narrower: the signals that indicate romance fraud are available well before payment instruction. When a PSP deploys a honeybot into a flagged conversation, the conversation length that makes romance fraud dangerous becomes the very thing that provides time to extract a scammer fingerprint and intervene.
How Romance Fraud Differs from Investment Scams, Operationally
Investment scam interception and romance fraud interception require different approaches, and conflating them produces inferior detection for both.
Investment scams are script-heavy. The scammer follows a process: introduce the opportunity, show fake returns, encourage reinvestment, engineer a withdrawal barrier. The conversation has a recognisable structure. Behavioural models trained on investment scam cadence do reasonably well because the pattern is consistent across operators, even across language barriers where translation is involved.
Romance fraud is structurally different. The conversation is designed to feel organic. A skilled romance scammer avoids script-like uniformity because uniformity signals inauthenticity to a victim who's been made to feel special. The emotional register of a romance fraud conversation — intimacy, crisis, urgency, reassurance — is harder to model from message cadence alone.
What does persist across romance fraud conversations: the identity construction patterns (military deployment, offshore oil rig, international business travel — specific scenarios that generate plausible absence and financial need), the platform migration requests (from dating app to WhatsApp to Telegram), the crisis narrative structure (sudden need arising just as the relationship has deepened), and the coaching behaviour that precedes the payment request.
Scammer fingerprinting for romance fraud therefore relies more on identity construction metadata and platform-switching signals than on message cadence alone. When a honeybot enters this type of engagement, the extraction goal is different: we're looking for the identity layer and the network patterns that connect this operator to known romance fraud infrastructure, not primarily the conversation script.
The Reimbursement Liability Angle
The PSR's mandatory reimbursement regime — now in force — has a specific interaction with romance fraud that PSP fraud ops teams should be tracking carefully.
The maximum reimbursement cap under the mandatory regime sits at £85,000 per claim. Romance fraud transfers frequently approach or exceed this threshold. A PSP handling significant romance fraud volume faces direct, predictable reimbursement liability on a category where transaction monitoring is demonstrably insufficient.
The PSR's rules include provisions around customer vulnerability and information. A PSP that receives a romance fraud reimbursement claim will face scrutiny on what warning mechanisms were in place, whether the customer received appropriate intervention at the payment stage, and whether the PSP had signals available that it failed to act on. We're not saying there's a simple test that absolves a PSP — the regulatory picture is more nuanced than that. But the direction of regulatory expectation is clear: passive reliance on SCA and transaction monitoring is not going to satisfy Consumer Duty or the PSR's reimbursement framework when the PSP could, in principle, have intervened earlier.
Where Interception Is Practical
The practical interception point for romance fraud sits in two places, and they serve different purposes.
The first is the flagged-conversation layer. When a customer's account shows signals consistent with an ongoing romance fraud engagement — unusual communication channel references in payment metadata, incoming payment patterns that match the initial "test transfer" structure, newly registered payee accounts — the PSP has a basis to inject a honeybot into the conversation stream at the customer's consent and instruction. The goal is fingerprint extraction and engagement stalling. This is early-stage interception: the aim is to characterise the operator before the large transfer request arrives.
The second is the payment-stage friction point. When a large payment to a new payee is initiated, targeted friction — a brief human-voice call, a structured delay with a counter-coaching message, a mandatory 24-hour hold on first-time transfers above a threshold — can break the urgency dynamic that romance fraud operators depend on. The scammer's ability to create a crisis narrative depends on the victim acting quickly. Any credible pause introduces cognitive space that a genuinely emotionally compromised victim can use to question the situation. This is not guaranteed to work; a victim who's been in an engagement for three months will often rationalise through a cooling-off intervention. But the data on 24-hour holds from PSPs that have implemented them indicates meaningful friction effects on this category.
The Victim Outcome Lens Matters for Ops Teams
One nuance worth naming explicitly: romance fraud victims frequently do not identify as fraud victims at the point of the initial transfer. They believe they are helping someone they care about. This creates a specific challenge for payment-stage intervention that investment scam intervention doesn't have to the same degree.
Telling a romance fraud victim at the payment stage that they may be being defrauded is not neutral. If wrong, it damages the victim's trust in their PSP and, depending on how it's handled, can cause genuine emotional harm. If right, it can also cause distress — the victim is being told that a relationship they believed was real is not.
The operational implication: romance fraud intervention messages need to be framed differently from investment fraud intervention messages. Language around "we want to make sure you're protected" and "take your time with this payment" is more appropriate than language that implies the victim has been fooled. Victim-centred intervention design isn't just ethically appropriate under Consumer Duty — it's more effective, because the defensive response to "you're being scammed" is often to accelerate the transfer to prove the bank wrong.
Building those messaging frameworks, and training fraud ops staff on the specific emotional dynamics of romance fraud victim engagement, is an area where PSPs often under-invest relative to the technical tooling. The conversation-layer tools are only as effective as the human escalation process they feed into.