Yes, when the dispute fits a clear reason and the evidence is complete. An AI agent can take the cardholder's dispute, check it falls within the scheme's time limits, gather the evidence and raise the chargeback at depth 3. Fraud claims, high-value disputes, contested cases and early credit to the cardholder outside policy stay with a named person.
Why is raising a chargeback suitable for an agent?
A chargeback is a request, not a verdict. The issuing bank asks the merchant's bank to return the money under the card scheme's rules, with a reason and evidence. The merchant can accept or contest. Nothing about raising it is final, and an issuer that raises a weak chargeback mostly loses time.
That makes the task a good match for an agent. Most of the work is structured: choosing the right reason, checking dates against the scheme's windows, collecting the receipt, the cancellation email, the delivery tracking. Getting those right first time is what wins cases. Our explainer on what a chargeback is covers the cycle.
What decides whether the agent can raise it?
The deciding factor is fit. Some disputes fit a reason cleanly: a duplicate charge, a subscription charged after a documented cancellation, goods not received well past the promised date with tracking showing no delivery. Others do not: "the product was worse than I expected", or a dispute that is really a claim that the card was used without permission.
That second case matters. If the cardholder says they did not make the transaction, it is an unauthorised transaction claim, not a merchant dispute. Different rules apply, including obligations to refund the customer in many markets, and it belongs in the fraud process. Treating it as a chargeback can leave the customer worse off.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Taking the dispute | Asks structured questions, depth 2 | Disputes the cardholder cannot describe |
| Separating fraud from merchant disputes | Routes unauthorised claims to fraud | The fraud decision |
| Checking time limits | Calculates them from transaction and expected dates | Late claims the bank may still pursue |
| Checking the cardholder contacted the merchant | Asks for evidence where policy requires | Waiving the requirement |
| Gathering evidence | Builds the pack against the reason's requirements | Nothing routine |
| Raising the chargeback | Submits at depth 3 for clear reasons within the value limit | Above the limit or unclear fit |
| Temporary credit to the cardholder | Applies the policy | Exceptions |
| Merchant contests the chargeback | Summarises the merchant's response | Deciding whether to pursue |
An example: a gym that kept charging
Take a cardholder who cancelled a gym membership in writing and kept the gym's email confirming it. Two months later, two more monthly charges of 45 have appeared. They message the bank.
The agent asks when they cancelled, whether they have the confirmation, and whether they contacted the gym about the later charges. The cardholder uploads the confirmation email and a message to the gym that went unanswered. The agent confirms both charges fall within the scheme's window, selects the reason for a cancelled recurring transaction, attaches the evidence and raises the chargeback. It tells the cardholder what happens next and how long the merchant has to respond.
The merchant does not contest, and the funds return. Had the gym produced a signed contract with a notice period, the agent would have summarised it and passed the case to a disputes analyst to decide whether to go further.
What controls make it safe?
- Reason requirements from the scheme rules. The evidence each reason needs is held in the policy library, and the agent checks the pack against it before raising.
- Fraud routing first. Every dispute is screened for "I didn't make this" before anything is raised.
- A value limit. Above the bank's threshold, raising is held for a person.
- Evidence provenance recorded. Cardholder uploads are declared evidence; merchant responses arrive through the scheme and carry its stamp.
- No pursuit after a contest. Second-cycle decisions are outside the agent's scope.
Merchants run the opposite process, defending chargebacks; see handling chargeback disputes with an AI agent. The evidence each side needs is listed in the chargeback dispute evidence checklist. The general limits are in AI agents in banking: what they may and may not do.
Where Pryme fits
In Pryme, raising a chargeback is certified on the issuer's own past disputes, including the ones that were really fraud. The scope gate fails any run where the agent raises a chargeback for an unauthorised claim. Pursuing a contested case is outside the role scope, so the agent's verdict on that action is refused and filed with the reason.
Questions people ask
Can the agent raise a chargeback after the time limit?
Usually not, because the scheme will reject it. The agent should tell the cardholder plainly and offer other routes, such as a complaint to the merchant, rather than raising a claim that cannot succeed.
Does the cardholder have to contact the merchant first?
Many reasons expect it, and many issuers require it by policy. The agent asks for evidence of contact where it is needed.
Can the agent withdraw a chargeback?
Yes, if the cardholder says the merchant has now refunded them directly, and the bank's process allows it. That avoids the cardholder being refunded twice.