Spot a customer being coached by a scammer mid-conversation

Fields

• Channel (chat or call transcript)
• The conversation so far, with speaker labels

Questions

1. Is the customer asking the bank to make, release or allow a payment, a transfer or a higher limit in this conversation? Yes / No

   Count a request to release a held payment, to send a new payment, to add or confirm a payee, to raise a daily or single payment limit, or to unblock a card or account so a payment can go through. Yes: The customer is asking for money to move, or for a limit or block to change so it can. No: The conversation is about something else, such as a balance, a lost card or a question, and no payment or limit is asked for.

2. Does someone else appear to be telling the customer what to do or say in this conversation? Yes / No

   Count "they said", "he told me to say", someone listening or talking on another line, the customer checking with someone before answering, or a named adviser, account manager or official who is guiding the payment. A relative helping at the customer's request, with no one else directing them, does not count. Yes: Someone other than the customer is directing the payment, the answers or the story. No: The customer speaks and decides for themselves, with no sign that anyone is steering them.

3. Do the customer's answers about the payment look scripted, changed or refused? Yes / No

   Count a reason for the payment that changes during the conversation, a reason the customer says they were told to give, answers that sound read out, and refusing to say what the payment is for ("why do you need to know, it's my money"). Do not count short, hesitant or unclear answers, a language the customer is less at home in, or a customer who is simply annoyed by the check. Yes: At least one answer about the payment or the payee looks scripted, changes, or is refused. No: The customer explains the payment in their own words and the explanation holds together.

4. Has the customer been told to keep the payment secret or not to trust the bank? Yes / No

   Count being told not to tell the bank, family or friends; being told the bank will try to stop the payment, is in on a fraud, or has staff involved in one; and being told to give the bank a false reason. Yes: The customer mentions at least one such instruction or belief. No: Nothing of this kind is said.

5. Does the customer mention installing an app or software for someone, or someone seeing or using their screen? Yes / No

   Count screen-sharing or remote access apps, an app installed at someone's request, someone logging in for the customer, or someone watching their screen or phone while they make the payment. Yes: The customer mentions something of this kind. No: Nothing of this kind is mentioned, or the customer says no one has access to their device.

6. Does the customer say the payment must happen now because of a deadline, a threat or a loss if they wait? Yes / No

   Count a stated deadline, a price window, a bonus or opportunity that will be lost, or a threat such as arrest, a fine or losing their money. An ordinary due date, such as rent due on the first of the month, does not count. Yes: The customer puts the payment under pressure of time or threat. No: No pressure of time or threat is stated.

7. Which scam story, if any, does the conversation most resemble? Choice

   Pick the one scam story named in the context that the conversation most resembles. Choose No scam story when the customer describes an ordinary payment. Judge the story, not the payee name.

   • No scam story An ordinary payment such as rent, a bill, a purchase from a known seller or money to family, with nothing that fits a scam story.
   • Bank or police impersonation Someone claiming to be from the bank, the police or another authority says the customer's money is at risk and must be moved to keep it safe.
   • Other impersonation Someone claiming to be from a tax office, a utility, a delivery firm or a phone company asks for payment of a bill, a fine or a fee.
   • Investment Crypto, a trading platform or a fund promising high or guaranteed returns, usually found online or through an adviser who made contact first.
   • Romance or friendship Money for someone the customer knows only online or has never met in person, often for travel, a medical bill or an emergency.
   • Purchase from a stranger Goods, a pet, tickets, a vehicle or a rental deposit paid by bank transfer to a seller found online, usually before seeing the item.
   • Invoice or bank detail change A supplier, builder, solicitor or landlord has sent new bank details for a payment the customer expected to make.
   • Advance fee A fee paid to release a loan, a prize, an inheritance, a parcel or a refund.
   • Job or task scam Paying to unlock earnings, commission or a job that comes from online tasks, reviews or orders.
   • Relative on a new number A message claiming to be from a child or another relative with a new phone number asks for money urgently.

8. Does the customer express any doubt or worry about whether the payment or the person behind it is genuine? Yes / No

   Count the customer wondering aloud, repeating a relative's or friend's warning, asking the agent whether the company or person is real, or saying something feels wrong. This is the moment the agent can use, so count it even when the customer goes on to insist on the payment. Yes: The customer voices at least one doubt, worry or question of this kind. No: The customer voices no doubt about the payment or the person behind it.

9. What scam warning has the agent given so far in the conversation? Choice

   Judge only what the agent has said so far. A tailored warning names the scam the conversation resembles and the signs of it in this customer's own story. Pick one option.

   • Tailored warning The agent names the kind of scam this looks like and explains the signs of it in the customer's own story, such as an adviser telling them what to say.
   • General warning only The agent gives only a general caution, such as "we see a lot of scams", without linking it to what the customer has described.
   • No warning The conversation shows signs of a scam and the agent has given no scam warning of any kind.
   • No warning needed The conversation shows no sign of a scam, and the agent has given no warning because none is called for.

10. How strongly does the conversation show the signs of a scam in progress? Scale

    Rate how clearly the conversation, taken as a whole, shows one of the scam stories or the signs of coaching named in the context. Rate what is said, not the person and not the amount.

    • None An ordinary conversation. Nothing fits a scam story or a sign of coaching.
    • Faint One weak sign, such as a vague reason or a passing mention of an online adviser, with no story behind it.
    • Clear A recognisable scam story, or several signs that point the same way.
    • Strong A clear scam story together with coaching, secrecy, a cover story, remote access or a deadline.

11. What should the agent do next in this conversation? Choice

    Suggest the next step for the agent, a trained person who decides. None of these steps releases, holds or returns a payment; the fraud specialist decides that with the customer. Pick one option.

    • Continue normally Nothing in the conversation calls for more than the usual checks.
    • Ask open questions Something does not add up. The agent uses the scam conversation script and asks open questions about the payment and who suggested it, in the customer's own words.
    • Bring in a fraud specialist now The conversation shows a scam story or coaching. The agent brings in a fraud specialist while the customer is still on the line.

Lens columns

payment_at_stake, payment_at_stake_probability, someone_directing, someone_directing_probability, cover_story, cover_story_probability, secrecy_or_distrust, secrecy_or_distrust_probability, remote_access, remote_access_probability, urgency, urgency_probability, scam_story, scam_story_probability, customer_doubt, customer_doubt_probability, warning_given, warning_given_probability, signal_strength, signal_strength_average, suggested_step, suggested_step_probability