Payments, scams and platform abuse
Your scam warnings pass review. Do they stop a customer who is mid-scam?
Under the UK reimbursement rules, a scam payment that gets through is now a direct loss for both the sending and the receiving firm. A warning screen only earns its keep if it stops the payment, and most are tested on calm, uncoached readers, not on someone a scammer has already prepared.
Built for Banks, neobanks, payment apps and wallets with APP (authorised push payment) scam exposure and reimbursement rules; Head of Fraud, Head of Scams, Payments Product
The problem in numbers
What the public record shows
Every figure links to the page it came from, with its date.
86%
of money lost to APP scams was returned to victims in the first three months
psr.org.uk, 2025-05
50:50
split of the reimbursement cost between the sending and the receiving firm
psr.org.uk, 2023-12
What you get
The deliverables
You get a lab where a scripted scammer plays against a copy of your real warning screens, a report on which warnings hold and which fail, and an optional pilot of a short explainer in the message you already send to suspected money mules.
A scripted scammer lab
Coaching scripts for your top scam types, such as safe account, investment and impersonation, played against a copy of your real warning screens in the order a customer sees them.
A hold-or-fail report
For each warning and each script: does the warning stop the payment, slow it, or get clicked through. Written so fraud, product and conduct teams can act on it.
Fixes ranked by effect
A short list of changes to wording, branching and friction, ordered by how much each one changed the lab result.
Optional mule-message pilot
A short explainer added to the message you already send to suspected money mules, with a before and after read on what those customers do next.
How it runs
3 weeks, start to finish
3 weeks
Week 1: scope and copy
You share a test copy of the warning screens, the payment purposes and the rules that decide when each warning appears. Amir agrees the scam types to test.
Week 2: lab runs
Scripted scammers play against the copy of your flow. Every warning variant is scored for every script.
Week 3: report and readout
A written report, a readout with your fraud, product and conduct teams, and the plan for the optional mule-message pilot.
Method
How the answer is produced
Scripts written from known scams
Each script predicts your warning and tells the customer how to get past it, the way real coaching does. The lab tests the warning against that, not against a calm reader.
Your screens, a test copy
The lab runs on a copy of your warning screens in the same order a customer sees them. No live customer is involved in the lab and nothing in production changes. The optional mule-message pilot is a separate step that runs only if you choose it.
Scored per warning
Results are kept per warning and per script, so you can tell which warning variants hold and which fail, instead of one overall loss rate.
Why Amir
Designed and led by someone who has built fraud models in production
The patents closest to this problem:
Amir is a named inventor on US 12,626,151 B2 (Method and system for testing a classification machine learning (ML) model of a tenant of a service provider, in a cloud-based environment, filed at NICE Actimize; granted).
The same idea of checking a decision system on realistic cases before anyone relies on it, here applied to a classification model in a shared cloud service.
View on Google PatentsAmir is a named inventor on US 12,541,721 B2 (Method for extreme class imbalance within fraud detection, filed at NICE Actimize; granted).
Scam and fraud events are a tiny share of traffic, so a warning's effect is hard to see in normal data. This patent addresses building models when the positive cases are extremely rare.
View on Google PatentsThese patents were filed at the companies named. They show the problems Amir has worked on; they are not Riskernel products.
Is this you
You probably need this if
- You show a warning before risky payments but have never tested it against a customer who has been coached by a scammer.
- Your scam losses are shared with another firm under the UK reimbursement rules, on the sending side or the receiving side.
- The warning a customer sees depends on the payment purpose they pick, and you cannot say what a coached customer would see.
- You cannot join the warning shown on a payment to the claim that followed it.
- The message you send to suspected money mules is written to inform, and you have never tested what the customer does next.
Questions
Before you reach out
Does this touch live customers?
Not the lab. It runs on a copy of your screens in a test environment, and no customer sees a changed screen unless you decide to ship a fix afterwards. The one exception is the optional mule-message pilot: if you choose it, your team adds the explainer to the live message you already send to suspected money mules, and the report compares what those customers do before and after.
What access do you need?
A test or staging copy of the payment journey, the text of each warning, the list of payment purposes and the rules that decide when each warning appears. No production customer data is needed.
How long does it take?
Three weeks from the day access to the test copy is ready.
Is the mule-message pilot required?
No. It is optional. It adds a short explainer to the message you already send to suspected money mules and measures what those customers do next.
Who should be in the room?
Whoever owns the warning copy, the thresholds that decide when it fires, and the claims data. If those are three teams, all three are welcome at the readout.
Scope Scam Warning Lab with Amir
Tell Amir what you run and what worries you. He replies with a short scope outline.