Jonathan Razi has done this before. He built CardX, a surcharge company, and sold it to Stax Payments in 2021. In June he launched his next act, an agentic service that automates the grind of fighting chargebacks for merchants. Payments Dive reported that his new company has raised $2.25 million and is already being resold through processors and facilitators.
The pitch is plain: higher win rates, lower labor costs. Razi says the tool does the dispute work that used to need a team of people, generating the evidence and the responses that get a chargeback reversed. For a merchant drowning in disputes, that is an easy yes.
The chargeback was already a fight. AI just armed both sides of it.
That is the part worth sitting with. This is not an isolated product. It is one end of an arms race that is about to reshape how disputes work.
The merchant side gets automated
Findustry AI, the company Razi founded in 2024, automates representment, the process of contesting a chargeback by assembling evidence that the transaction was legitimate. Payments Dive reported it launched its public chargeback service this June, with clients including Cleveland-based Kurv, which resells the tool to merchants, and Indianapolis facilitator AllPaid, which serves municipal and county clients.
The economics are obvious. Representment is tedious, deadline-driven, and expensive to staff. An agent that drafts the rebuttal and pulls the supporting records turns a back-office cost center into a button. Razi told Payments Dive the result is an increased win rate and labor savings, and that the company does not need more funding to scale it.
The other side is automating too
Here is what makes it an arms race. Issuers and their fraud vendors are not standing still. The same generative AI that writes a merchant's dispute defense can write a cardholder's claim, and banks are already using AI to triage, score, and contest at volume. We covered this dynamic in our look at the AI fraud paradox: the technology that defends the system is the same technology straining it.
When both sides automate, the dispute stops being a human judgment and becomes a contest between models. The merchant's agent argues the charge was valid. The issuer's agent argues it was not. Volume goes up, cost per dispute goes down, and the signal a chargeback was supposed to carry, that something actually went wrong, gets noisier.
The gap nobody has closed
None of this answers the question agentic commerce keeps raising: who owns the dispute when an agent made the purchase?
A human buys the wrong thing and the liability chain is well understood. An agent buys the wrong thing, inside the scope its owner authorized, and the chain breaks. Was it the cardholder who set the budget, the merchant who sold to a bot, the issuer who approved the token, or the developer whose model misread the page? We flagged this missing piece in our analysis of the agentic dispute layer, and automated chargeback tools make it sharper, not softer.
Automating representment assumes a dispute has a knowable answer. Agentic purchases blur exactly that. A tool optimized to win chargebacks does not care whether the underlying purchase was a genuine error by an autonomous agent. It cares about the win rate.
What breaks
The risk is that both sides optimize for winning rather than for truth. If a merchant's agent gets very good at reversing chargebacks and an issuer's agent gets very good at filing them, the dispute system measures who has the better model, not who was wronged. We described the early version of this in our coverage of the agentic dispute crisis.
That is a problem for everyone who relies on chargeback data as a signal. Card networks use dispute rates to flag bad merchants. Issuers use them to spot compromised cards. If those rates are increasingly the output of dueling automation, they start to measure something else.
What to watch
Two things. First, whether the networks update dispute rules for agent-initiated purchases, because the current rules assume a human cardholder. Second, whether win-rate guarantees become a selling point, which is the tell that the system is being gamed rather than fixed.
The tools to fight disputes are racing ahead of the rules that govern them. That is good business for the people selling the tools. It is an open question for everyone settling the bill.
If both the merchant and the bank let an AI fight the chargeback, who is left to decide what actually happened?
Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.