Automating the rebuttal does not win the chargeback

Chargeflow raised a $35 million Series A on 18 November 2025, taking total funding to $49 million, and says its software powers more than 15,000 merchants, beats friendly fraud at “up to 80%,” and is built to “prevent nine out of ten disputes before they even occur” (Chargeflow Series A announcement). What the round capitalises is the disappearance of a job. The merchant’s reply to a chargeback, the representment, used to be a person gathering delivery confirmation, login records and correspondence and writing them into an argument before a deadline. It is now a model that gathers and drafts, sold on the promise that the merchant never has to read the result. The pitch, and the valuation behind it, rests on an equation that does not hold: that a rebuttal produced cheaply and instantly is the same as a rebuttal that wins.

A representment is an argument, and arguments are not won by volume of evidence. They are won by selecting the two or three facts that actually rebut the specific reason code under dispute and presenting them so an issuer’s decision, increasingly an automated one, resolves in the merchant’s favour. Justt describes a platform that analyses “over 500 data points from multiple sources to create tailored, high-quality arguments for each case” (Justt Series C announcement), and the figure is meant to impress, but 500 data points are inert until something decides which of them matters for this dispute and frames them as a case. Automated data is only ever as good as its presentation in the representment, and presentation is the part the data pipeline does not do. A rebuttal that returns the full data dump against a 10.4 fraud claim is weaker than one that surfaces the single prior undisputed transaction sharing a device fingerprint and shipping address and argues from it, even though the second uses a fraction of the inputs.

The economics of the automated model push in exactly the wrong direction on this. A vendor billing the merchant only when it wins is optimising for the cases it can win cheaply at scale, which are the data-provable ones where assembling the evidence is most of the work, and those are precisely the cases the networks are moving to resolve without a representment at all. The contested disputes, where the evidence is ambiguous and the outcome turns on how persuasively it is argued, are the expensive, low-throughput cases an automation business is structured to avoid. The merchant is sold a win rate that is highest exactly where the value of winning is about to fall, and lowest exactly where it is rising.

The behaviour underneath is getting larger, which raises the value of a built argument rather than lowering it. The Merchant Risk Council’s 2026 report finds first-party misuse, disputes a cardholder files against a charge they genuinely made, rising across the surveyed merchant base, with a majority reporting it worsening (MRC 2026 Global Payments and Fraud Report). Visa disclosed processing 106 million disputes globally in 2025, up 35% from 2019 (American Banker, November 2025), and a consumer who can have a model draft a plausible dispute and pick the paying reason code faces the same collapse in cost the merchant’s vendor enjoys on the other side. When both sides automate and the volume climbs, every rebuttal run on the same models against the same data converges toward the same template, and an adjudicator facing a flood of near-identical filings discounts the generic one. Differentiation moves to the rebuttal that does not look automated.

The networks have read the same trend and are taking the provable cases for themselves, which is what hollows out the data-assembly business. Visa’s Compelling Evidence 3.0, through its auto-qualification path, checks a merchant’s prior-transaction history against a disputed charge in the pre-dispute window and qualifies the evidence automatically, deflecting a fraud claim that meets the data test before it becomes a chargeback anyone argues (cside on CE 3.0 auto-qualification through Visa Secure; Visa CE 3.0 merchant readiness guide). Mastercard’s First Party Trust programme, expanded in June 2025, does the parallel job, passing merchant transaction data to the issuer to identify and decline a first-party fraud claim at source (Mastercard First Party Trust announcement). Both automate the data-provable rebuttal, which is the exact service the automation vendors sell. The pure-data representment is being absorbed into the network rail, and what is left outside it is the case that does not auto-qualify.

The pre-dispute rail makes the split explicit and prices it. Verifi, which Visa owns, runs Rapid Dispute Resolution, an automated system that refunds a qualifying dispute under merchant-set rules before it becomes a chargeback, at a reported $10 to $19 per resolved dispute and with 97% of US Visa issuers connected (chargebackstop on Verifi RDR; Visa dispute-resolution overview). The easy, data-clear disputes get resolved by the network for a fixed fee, and the tightening of monitoring programmes such as the 2026 Visa Acquirer Monitoring Program raises the cost of any dispute a merchant neither prevents nor wins outright (Basis Theory on the 2026 VAMP update). The disputes that remain for a full representment are the contested ones, the cases where a recovered chargeback genuinely matters and where the outcome depends on the quality of the argument rather than the completeness of the file. That residue is where representment as a craft earns its keep, and it is the part automation is worst at.

There is a serious objection here, and it is that the human-letter advantage is temporary. Generative models will keep improving and will soon write bespoke, persuasive rebuttals tuned to the reason code and the issuer, at which point the craft argument is automated too and scale beats craft on cost. The vendor win rates may be real today, and a merchant choosing between a cheap automated filing and an expensive manual one will reasonably take the cheap one while it wins. On this view the customised letter is a transitional advantage that the next model release erases.

Concede that models will narrow the gap, and the position still holds, because the adjudication is moving toward verifiable evidence and pre-dispute data integration, not toward better prose. CE 3.0 and First Party Trust reward the merchant who has the right data, matched precisely to the dispute and submitted before it escalates, and that is a question of evidence quality and how it is presented, not of how many tokens the rebuttal runs to. An automated filing that contests everything to maximise a success fee is optimised for throughput, and throughput is the commodity the networks are pricing at $10 to $19. The defensible business is the one that competes on the specificity and credibility of the evidence and on getting it in front of the right automated rule at the right moment, which a model can assist but which is not the same as filing more. Automation buys a merchant participation in the dispute, cheaply. It does not buy the win, and on the cases that pay it is being out-competed by the network on one side and by argument on the other.

For anyone underwriting a chargeback-management or representment business, an issuer’s dispute function, or a network’s services line through 2027, the diligence question is which side of that split the asset sits on. A vendor whose edge is automated data assembly and a success-fee win rate is selling a service the networks are internalising for a fixed fee, and its advertised numbers are a pre-saturation arbitrage on a labour cost that is disappearing, so the win rate is the wrong thing to underwrite. The durable position is evidence quality, pre-dispute data integration, and an argument matched to the specific dispute, aligned with the merchant rather than with the vendor’s fee. Press whether the asset’s representments win the contested cases that still require a built argument, or only the provable ones the network is about to resolve without it, because the first is a business and the second is a line item Visa already prices.