Scam reimbursement needs a person

Deciding a scam reimbursement claim means judging a person at their worst moment. Did they ignore a warning, or were they coached to? Were they vulnerable, and did that shape what they did? Was their account credible? Those are judgements about behaviour and circumstances, with large consequences for someone already harmed. An agent can prepare the claim thoroughly. A person should decide it.

What does a reimbursement decision actually weigh?

When a customer reports that they were tricked into sending money to a fraudster, the bank has to decide whether to reimburse them. In some markets, rules require payment firms to reimburse many victims of authorised push payment scams, with exceptions that turn on the customer's conduct and circumstances. Elsewhere, reimbursement is at the bank's discretion or set by its own policy.

Either way, the decision rests on questions that are not mechanical.

None of these has a clean rule. Each involves weighing the person's situation against standards of care, often from a conversation, and the answer matters a great deal to them.

Why should an agent not make the decision?

Three reasons, each sufficient on its own.

The harm of a wrong decline is severe. A customer who loses savings to a scam, and is then told the bank will not reimburse them, faces a second blow. Some will pursue a complaint; many will not. A decision that may cost someone a large share of their savings should have a person's name on it.

Vulnerability is hard to read and easy to miss. Signs of vulnerability can be subtle and may emerge only in conversation. A customer who seems careless in the transaction data may turn out to have been newly bereaved and targeted for it. Recognising that, and weighing it, is a human judgement.

The agent may be part of the story. If an agent gave the customer warnings during the payment, or answered their questions about a name mismatch, the adequacy of those interactions is part of the assessment. An agent should not judge whether its own warnings were good enough.

What can the agent carry?

A great deal, and doing it well shortens the time a victim waits.

• Taking the report with care, in plain language, without implying blame. The guide on whether an AI agent can handle a scam victim's report covers the conversation.
• Containment: contacting the receiving bank quickly to try to freeze remaining funds, which is time-critical.
• Building the timeline: every payment, every warning shown, every contact with the customer, from the bank's own records.
• Gathering the customer's evidence: messages from the scammer, adverts, documents, organised and dated.
• Drafting the assessment against the bank's framework, setting out the facts for each question and flagging where judgement is needed. A scam reimbursement claim assessment template gives the structure.
• Drafting the decision letter for the assessor to edit and approve.

The process end to end is in how to handle push payment scam reimbursement claims with an AI agent.

What makes the agent's preparation fair?

Preparation can tilt a decision as much as the decision itself. An assessment pack that leads with the customer's mistakes and buries the scammer's tactics invites a decline. Four rules keep it balanced.

1. Present the scammer's methods alongside the customer's actions, so the assessor sees what the customer was up against.
2. Surface every possible vulnerability indicator, even uncertain ones, for the assessor to weigh.
3. Separate facts from the bank's records, which are sealed, from the customer's account, which is declared, without treating declared as doubtful.
4. Do not recommend an outcome on the questions that need judgement. State the facts and leave the weighing to the assessor.

The fourth rule is a deliberate limit. An agent recommendation on "did the customer act with appropriate care?" anchors the assessor. On mechanical questions, such as whether the payment falls within the scope of the scheme, a recommendation is fine.

How does this fit with the bank's other judgements?

Scam reimbursement belongs with a small set of bank decisions that weigh a person's circumstances: enhanced due diligence, customer exits, and treatment of customers in financial difficulty. In each, agents gather and prepare, and people decide. Prevention is the other side: an agent's role in stopping an authorised push payment scam before money leaves is larger, because holding a payment is reversible and reimbursement is not needed if the scam fails.

Where Pryme fits

In Pryme, taking reports, containment and evidence assembly can each earn their own depth on your own claims. Recording a reimbursement decision is held for a named assessor. The facts, their provenance, and the assessor's reasons are filed together, so every decision can be reviewed if the customer complains. See AI agents in banking: what they may and may not do.