Chargebacks: issuer fraud verdicts and the impact of agentic commerce

Chaira Mekkaoui, Associate Director at Redbridge Debt and Treasury Advisory, analyses banks' approach to determining whether a merchant's customer is fraudulent during the chargeback defence procedure, with a particular focus on new AI-related use cases.

Fraudster or not? AI reopened aspects issuers automated

In Europe, an issuer's fraud verdict is changing from human judgement to data-enabled decisions. Agentic commerce is about to force a question other than who pays when an AI agent buys: how can an issuer verify a purchase no human consciously made?

When a merchant must defend a chargeback, it can feel like arguing before a judge. By the time the dispute arrives, the issuer has catalogued the transaction as fraudulent and refunded the customer. An issuer prioritises cardholder satisfaction, so representment challenges a decision reached under incentives that rarely favour the merchant.

The decision behind the question for each dispute, ‘is this customer a fraudster?’, although solved by data thanks to automation, is now more difficult because of agentic commerce. How can an issuer authenticate a part-human, part-machine customer?

Two approaches for the same chargeback decision

Chargeback determination used to be discretionary: the cardholder's word, the issuer's own fraud signals, and a tendency to refund. Now, the process is becoming deterministic, testing whether the merchant's data matches a trusted footprint.

In the US, schemes drive this shift. Visa's Compelling Evidence 3.0 lets a merchant overturn a fraud dispute by producing prior undisputed transactions that share identifiers such as device ID or IP address. Mastercard's First-Party Trust does the equivalent.

Europe did the same with a different approach. Under PSD2, Strong Customer Authentication (SCA) settles much of the fraud question up front: an authenticated transaction transfers fraud liability to the issuer before a dispute exists. However, removing a category exposes another. As Gabriel Lucas covered in a previous article, that fall in point-of-payment fraud is precisely why policy abuse has risen so sharply in Europe: genuine customers exploit legitimate rules with valid identities and real payment methods, which tools built for stolen cards never flag.

The same dynamic reaches the dispute itself as first-party, or ‘friendly fraud’. Here, its structure is harder for the issuer, because SCA offers no leverage against a customer who authenticated the payment themselves. Europe only changed the nature of the fight against fraud.

What does proving fraud mean now?

This changed approach impacts defence mechanics. For an authenticated European transaction, a merchant rarely needs to argue whether the shopper was really the cardholder. Authentication generally addresses this, and the purely fraudulent chargeback is largely blocked at source.

The remaining dispute is the first-party one: a real, authenticated customer received the goods and disputes anyway. In this case, the issuer is no longer weighing identity but intent, and the evidence comes from behavioural data: device continuity, prior undisputed history, and delivery confirmation. Determining fraud changed from identifying who transacted to ensuring the behavioural pattern holds. It is the same data-matching logic card schemes hard-code, received from the opposite end of the transaction: the incoming Payment Services Regulation (PSR) will prompt issuers to monitor the device and behavioural signals schemes already treat as proof. Redbridge has traced this approach in its work on European card-scheme regulation.

Who will be impacted by the new requirements for fraud defence and chargebacks?

These procedures are not mere compliance footnotes, as they rewrite who wins disputes and who carries the cost.

For merchants, a well-instrumented transaction is defensible. The data currently deciding a dispute’s validity (device, identity, authentication, delivery) must be captured before the sale, not collected after the chargeback. This helps data-rich large platforms and exposes smaller merchants forced to defend disputes without the signals needed to decide. Instead of arguing representments, the fraud-fighting process changes to engineering qualification.

For PSPs and gateways, this is both an opportunity and liability. They are becoming the orchestration layer, capturing and structuring evidence, running SCA and its exemptions, and qualifying transactions for the schemes' programmes. The incoming PSD3/PSR regulation raises the stakes: it will enable delegated authentication and hold whoever performs it liable when it fails. The authenticator owns the risk, and PSPs are taking on issuer-like exposure.

For fraud and chargeback specialists, the decision moves fastest of all. Once determination is automated and evidence-gated, manual representment value decreases, and the process moves to earlier stages: pre-dispute deflection, alert networks, and keeping merchants qualified and beneath monitoring thresholds. Fighting disputes is turning into stopping them from being filed.

The agentic challenge

With agentic payments, the model breaks. When an AI agent shops and checks out for a consumer, the payment uses existing rails and, to the issuer, it looks like any card-not-present order, assessed by controls written to read human intent. This works only until the issuer has to authenticate the buyer.

The issuers’ challenge with agentic commerce is not identifying the payer. Schemes are already addressing this: they bind consent and agent identity into tokens, and default fraud liability to the issuer for validly issued credentials, with the consumer's dispute rights intact. The complexity lies in how an issuer performs its control when no human is present to authenticate. SCA assumes a person is authenticating; however, an agent acting on delegated authority fits that model awkwardly. The issuer's determination must therefore expand from identifying whether a cardholder is genuine to whether an agent is validly mandated, legitimate, and acting within scope. The reason codes were never written for a chain-of-authority test, and PSD3/PSR delegated-authentication rules will have to address this. Ultimately, determination does not depend on data only, and it must now identify the machine's authority, not the human's identity.

The irony

For a decade, the industry focused on removing judgement from the dispute, to make ‘fraudster or not’ a matter of data. Agentic commerce puts judgement back in, with a higher complexity: no longer asking ‘did this person defraud?’ but ‘can the issuer verify what a machine did on someone's behalf?’

The issuers, schemes, and PSPs that resolve this first will set the trust terms for the next decade of European payments.

About Chaira Mekkaoui

With 15+ years of payments experience, Chaira joined Redbridge in 2022 to help bring the payment practice to the next level. After a few years working as a payment consultant, she also worked at Visa for 10 years as a Fraud Manager and Business Analyst.

 

 

About Redbridge

Redbridge Debt and Treasury Advisory is a leading financial management partner to corporations around the globe. It is committed to providing each client with all the information required to make the best decisions and optimise their financial performance. Redbridge’s teams are in Houston, New York, Chicago, Paris, Geneva, and London.