Chargebacks Are Beautiful
How AI and Smart Operations Help Multi-PSP Merchants Win More Disputes
Most merchant conversations I have about chargebacks open with frustration. Dispute volume is too high. The team handling representments is too small. Each PSP has its own portal. The whole thing feels like a tax the business pays for the privilege of selling online.
I want to argue something different. Chargebacks, as a mechanism, are one of the better things about card payments. The pain we feel as merchants is real, but it sits in the operational layer above the mechanism, not in the mechanism itself. The merchants who internalize that distinction tend to be the ones who win more disputes, keep more customers, and build a dispute operation that gets stronger over time rather than more expensive.
Why chargebacks are beautiful
Let me state four things, briefly, in defense of chargebacks.
They create trust. Card schemes like to think it is their brand that creates trust. In reality, it's the knowledge that if something goes wrong they can get their money back. A consumer can buy from any online shop with their Visa or Mastercard and know they can get reimbursed if something bad happens. That trust is the reason ecommerce works at the scale it does. Take it away, and cart abandonment goes up everywhere.
They offer a fair process. Both merchants and customers get a chance to present their case. For merchants, that is especially nice to have when you compare it to a PayPal dispute. Or, god forbid, a SEPA Direct Debit return. The chargeback process, painful as it can be, is more balanced than most of the alternatives in the wider payments world.
They create a stable ecosystem. Chargebacks are unified and global. Reason codes are standardized. Timeframes are standardized. They drastically reduce incentives for merchant scammers, because consumers have a credible remedy. And they push every merchant in the system to improve the boring fundamentals: billing descriptors, delivery notifications, refund policies, a working customer hotline (operations that should exist anyway).
They are increasingly digital. Disputes that used to involve faxed evidence and weeks of back-and-forth can now be automated into a merchant's workflow. The technical surface for managing them has improved dramatically in the last few years.
Where the pain actually sits
And yet chargebacks are very, very expensive for merchants. Most of that expense has almost nothing to do with the mechanism itself. It comes from how dispute operations are run, and especially from the friction of running them across multiple PSPs at once.
Each acquirer exposes a different portal, a different evidence format, a different set of deadlines, a different way of reporting reason codes. Most teams handle that by stitching things together with shared inboxes and spreadsheets. As a result, cases get missed. Deadlines get blown. Win-rate data is impossible to compare across acquirers because each PSP defines it slightly differently.
What follows are five operational moves I see consistently separating merchants who handle this well from the ones who do not. AI shows up in several of them, though usually not where the marketing decks suggest.
1. Win the dispute before it becomes a dispute
Most representments are lost in the weeks before the chargeback is filed. The merchant descriptor is unclear, so the cardholder does not recognize the charge on their statement. Customer service replies in three days when the customer has already called their issuing bank.
This is the unglamorous side of dispute management, and the one with the highest return on effort. Clear descriptors that pair the merchant name with transaction context. A refund flow that takes a customer less effort than calling their issuer. Proactive transaction alerts for high-risk cohorts. Fraud signals routed back into the checkout, not only into post-authorization review.
The merchants with low chargeback rates almost always have boring upstream hygiene before they have any kind of dispute platform.
2. Centralize case management across PSPs
A multi-PSP setup is good for routing, cost, and resilience. It is brutal for dispute operations.
Every acquirer has a different portal, a different file format for evidence, a different representment deadline, and a different way of communicating reason codes. A merchant running four PSPs is effectively running four parallel dispute teams. Cases get missed because someone forgot to check a portal. Win-rate data is impossible to compare because every PSP reports it differently.
The fix is gloriously unsexy: pull every case from every acquirer and payment method into a single workspace, with a single deadline view, a single evidence template per reason code, and a single source of win-rate truth.
Chargebacks are one of the places in payment operations where the potential of unified data really shines.
3. Use AI where it removes a decision a human should not be making
AI is genuinely useful in dispute management for three things, in order of maturity.
First, generating representment narratives from structured order and transaction data, with the right evidence attached. A representment that took an analyst forty minutes to assemble can become a five-minute review of a draft.
Second, clustering disputes by likely root cause, so a spike in "product not received" complaints surfaces as a fulfillment issue before the win rate moves.
Third, predicting which disputes are worth fighting, given reason code, transaction value, customer history, and historical win rate for similar cases. Fighting everything is expensive. Fighting nothing leaves money on the table.
What AI is not useful for, at least not yet, is replacing the judgment that decides what your dispute policy should be. A model can draft a representment. It cannot decide whether you want to fight friendly fraud aggressively at the risk of customer churn, or absorb the cost to protect the relationship. That call belongs to the business.
One pattern in our industry deserves to be called out. When a vendor charges a percentage of recovered funds, the math rarely works in the merchant's favor, because they end up paying for representments they would have won anyway. A flat fee per case, or a platform fee, aligns incentives much better.
4. Build evidence templates per reason code, not per dispute
When an analyst writes a representment from scratch, two things happen. The quality is inconsistent, and the cycle time is long. The fix is to map every reason code your business sees regularly to a standard evidence pack: which order fields, which delivery confirmation, which prior customer communications, which fraud signals from the original authorization.
Once that mapping exists, AI can populate it in seconds. Even without AI, a structured template lifts win rates because it forces the team to attach the evidence the issuer actually needs, in the order they need to see it.
This is the single highest-leverage operational change that merchants can make. It is also one of the easiest to underinvest in, because it looks like documentation work rather than product work.
5. Treat win rate as a feedback loop, not a metric
Most merchants track win rate as a number in a monthly report. The merchants who win more disputes treat it as a feedback loop into the rest of the payments stack.
A rising win rate on "fraudulent transaction" disputes might mean the fraud engine is over-blocking and customers are filing false claims. A falling win rate on "product not received" disputes might mean a courier partner has degraded. A sudden spike in disputes from one issuer might point to a tokenization or 3DS configuration issue worth investigating before it gets expensive.
The data is in the cases. Few teams have it close enough to act on weekly.
A closing thought
Chargebacks are not the problem worth solving. The mechanism works. It creates the trust that lets ecommerce function. It gives both sides a fair process. It pushes every merchant in the system to do the basic operational work better.
The opportunity for merchants is in how we respond. The merchants who treat dispute operations as a serious product problem (centralized, templated, AI-supported where it earns its keep, feeding insight back into fraud and fulfillment) end up winning more disputes, keeping more customers, and learning more from the data along the way.
The shortest version of all of this: the dispute layer is a product problem, not a back-office cost. The merchants and platforms who treat it that way are the ones who will pull ahead.
Chargebacks are beautiful. The operations around them are what we get to make better.
About the author
Alexandru Dorobantu is Head of Payment Ecosystem at Payrails. He joined Payrails in 2025 from DeliveryHero, and now leads the team building Payrails' dispute and chargeback capabilities.
About Payrails
Payrails helps businesses use payments as the foundation for optimizing growth and global performance. Based across hubs in Berlin, Dubai, Dublin, and Cairo, Payrails works with global merchants to orchestrate, vault, route, reconcile, and manage disputes across multiple payment service providers.
