O comércio agênico e seu impacto sobre a fraude e o “ Chargebacks ”: reflexões de Frank Frantz

Perspectivas de Frank Frantz, executivo de desenvolvimento de negócios da Chargeflow

Agentic commerce is shopping in which an AI agent searches, chooses, and pays on a customer's behalf using delegated payment credentials. For merchants it separates the person who wants the item from the actor at checkout, which changes what you can prove about who authorized an order once it is disputed.

At Money20/20 USA 2025, the theme kept resurfacing in the most useful conversations: agentic commerce is here, and it is already reshaping fraud and chargebacks. In his interview on Bankadelic Live, Chargeflow's Frank Frantz argued that the biggest changes in ecommerce are not happening on storefronts. They are happening in the invisible layers of automation underneath them.

This page covers his four takeaways, the independent data that tests them, and where to go next for liability, evidence, regulation, and fraud prevention.

Sources: Mastercard and Datos Insights, The chargeback window of opportunity (March 2025); Adobe Analytics, January 2026.

1. O atrito com os clientes está, discretamente, provocando uma crise de “ Chargeback ”

Frank's first point: the line between customer experience and chargeback risk is thinner than most merchants budget for.

“Quando os clientes não conseguem obter um atendimento rápido e eficaz, eles ignoram completamente o processo e vão direto ao banco.”

• Slow support leads to frustration
• Frustration leads to disputes
• Disputes become chargebacks

These are not professional fraudsters. They are real customers who feel boxed out of service channels and take the shortest path to a refund, a pattern explained in the psychology of chargebacks. Support failures are a large indirect driver of disputes, and it is cheaper to invest in support and the ways to prevent chargeback fraud than to fight each dispute afterward.

2. A fraude por pessoas próximas não é um caso isolado. É a regra.

On the Bankadelic stage, Frank reminded listeners that the chargeback mix is not what most merchants assume:

“Oitenta por cento dos sites chargebacks são, na verdade, fraudulentos.”

He did not mean stolen cards or hacked accounts. He meant cardholders who received the goods and dispute the charge anyway, the pattern known as friendly fraud.

The exact share depends on who does the counting. Chargeflow's own State of Chargebacks analysis, covered in 80% of chargebacks due to friendly fraud, put friendly fraud at roughly 8 in 10 disputes across its merchant base. In the Mastercard and Datos Insights survey, merchants classified about 45% of their chargeback volume as fraudulent (first-party and third-party combined), while issuers classified 72% of disputes as fraudulent. Treat the figures as a range. First-party misuse is large enough that it cannot be planned for as an edge case.

Money20/20 sessions echoed the same shift: away from purely criminal fraud and toward first-party misuse, policy abuse, and “digital shoplifting,” a term Frank used in his interview. Merchants across verticals reported:

• Maiores volumes de “ Disputa ”
• Mais alegações do tipo “não fui eu”
• Customers using chargebacks as a convenience button
• Maior pressão operacional sobre equipes pequenas

Two controls work on this pattern before a formal chargeback lands. Chargeback alerts flag a dispute early enough to refund the order and keep it out of your ratio. For disputes that do land, Visa's Compelling Evidence 3.0 lets two prior undisputed transactions with matching customer data defeat a first-party fraud claim filed under dispute condition 10.4.

3. O comércio agênico é a nova fronteira (e a nova camada de risco)

The strongest trend on the show floor was a question rather than a booth: what happens when AI agents start making purchases on behalf of customers?

Frank put the shift plainly:

“Platforms like [AI assistants] are now letting customers research and buy in the same place, with their stored payment credentials.”

The traffic data points the same way. The Adobe Analytics figures above show AI-referred shoppers arriving in volume and converting better than other traffic, which is why merchants will accept agent-placed orders. The same shift introduces new challenges:

• An AI agent may buy the wrong item
• Os clientes podem esquecer o que delegaram
• Os fraudadores podem imitar padrões automatizados
• Lojistas pode perder a percepção da intenção
• Traditional fraud tooling can't read agent-driven behavior

AI-initiated transactions also blur accountability, which raises three questions: who authorized the purchase, who is the merchant of record, and who is liable when an AI makes a mistake. The liability answer is not settled. AI agent chargeback liability sets out who pays in each scenario, and who is liable when AI agents shop for your customers covers what counts as an authorized agent purchase.

The practical difference shows up in the evidence you hold. Most disputes are still filed under existing card network rules, so the question is what each side can prove:

Sources: Mastercard, agentic token framework; Visa, Trusted Agent Protocol announcement (October 14, 2025); Visa, agentic payments report (July 2026). Visa Intelligent Commerce Connect, announced April 8, 2026, remained in pilot with select partners at announcement.

One consequence deserves attention. Compelling Evidence 3.0 requires matching data elements across the disputed order and two prior undisputed transactions between 120 and 365 days old, and at least one matching element must be the IP address or device ID. If an agent places the order from its own infrastructure, those two signals may describe the agent rather than your customer, which leaves account ID and delivery address as the weaker matches. The agentic commerce chargebacks evidence playbook covers what to capture instead.

4. Repeat Disputers Look Like Good Customers: Where Chargeflow Prevent Fits

Frank was asked whether Chargeflow tracks repeat abusers across its network. His answer:

“Lançamos um produto no mês passado que faz exatamente isso.”

That product is Chargeflow Prevent. It scans orders after checkout and before fulfillment, between the transaction and the shipment in the order-to-cash process, and draws on a network of 20,000+ merchants. A customer who is new to your store can still carry a dispute history from others. Prevent cancels high-risk orders, requests verification on suspicious ones, and approves the rest, following rules you configure, with a false positive rate under 0.1%.

Frank's reasoning for why this matters as agents arrive:

• Agentic commerce increases transaction volume, and higher volume gives abusers more openings
• Atualmente, os consumidores Disputa realizam compras iniciadas por Automação
• Chargeback abusers slip past fraud prevention because they look exactly like good customers: same card, same identity, same behavior

A IA está transformando o processo de pagamento. E o “ Chargeback ”.

Frank's direction of travel: fraud is getting subtler, chargebacks are becoming more behavioral, and customers are delegating more decisions, so protection has to work after the purchase as well as at checkout. Digital goods make this harder still, since digital goods chargebacks already lack the proof of delivery physical orders provide.

What to do now:

• Write agent terms into your policies. Deloitte's legal-risk guidance recommends updating terms and conditions for AI agent transactions, requiring human confirmation for high-value purchases or initial agent setup, and defining clear authority limits.
• Watch your ratios. More disputes push you toward monitoring programs: Visa VAMP flags merchants as Excessive at a 1.5% ratio, and the Mastercard Excessive Chargeback Program starts at a 1.5% ratio with 100 chargebacks.
• Close the loop on disputes. Mature chargeback management blends prevention, alerts, and representment, and smart chargeback recovery turns lost disputes back into revenue.
• Treat agent orders as card-not-present orders. They carry the same exposure as card-not-present fraud, plus the authorization questions above.

The specialist guides below go deeper on each angle:

Chargeflow covers the three stages: Alerts before a dispute becomes a chargeback, Prevent between checkout and fulfillment, and Automation for evidence and submission on the disputes that still land.

Perguntas frequentes

What is agentic commerce?

Agentic commerce is online shopping in which an AI agent researches, selects, and pays for items on a customer's behalf using delegated credentials, instead of the customer completing checkout personally. For merchants, it means orders can arrive from an agent acting within limits the customer set earlier.

Is agentic commerce real or still hype?

It is real but early. Adobe Analytics recorded generative AI referrals to US retail sites rising nearly eightfold year over year in the 2025 holiday season, Mastercard unveiled Agent Pay in April 2025, and Visa released its Trusted Agent Protocol in October 2025. Visa's Intelligent Commerce Connect was still in pilot with select partners when announced in April 2026, so volumes and dispute rules are still forming.

How does agentic commerce affect chargebacks?

It changes the dispute story more than the dispute mechanics. Customers can claim they never approved what an agent bought or that it bought the wrong item, and the usual evidence of device, IP address, and browsing session may describe the agent instead of the customer. More disputes also raise the ratios that card network monitoring programs track. Mastercard says agentic token intent data can serve as an audit trail to help resolve cardholder disputes.

What fraud risks does agentic commerce create for merchants?

The main risks are fraudsters mimicking automated agent patterns, stolen credentials used through agent flows, blurred authorization when a customer forgets what they delegated, friendly fraud claims against delegated purchases, and bot defenses that block legitimate agents. Agent orders are card-not-present transactions, so the controls in an ecommerce fraud prevention program still apply.

Are chargebacks considered fraud?

Not always. A chargeback is the reversal of a card payment after a cardholder dispute (see what is a chargeback), and the cause can be third-party fraud, first-party misuse, or merchant error. Mastercard and Datos Insights found that issuers label a much larger share of disputes as fraud than merchants do, so the answer depends on who is counting.