Every conversation about "friendly fraud" eventually arrives at the same framing: bad actors gaming a system that favours consumers. That framing isn't wrong, but it's incomplete. Most disputes that get labelled "friendly fraud" aren't acts of dishonesty. They're the predictable output of a consumer trying to solve a problem through the easiest channel available to them.

The scale of the problem isn't in dispute, even if the exact figure varies by source and year. Visa estimated in 2022 that friendly fraud accounted for 75% of total chargeback volume. Chargeback911 put the figure at 86% in 2024. The two numbers come from different methodologies and different points in time, so they shouldn't be read as a single trend line — but directionally, they point the same way: on any reasonable estimate, the clear majority of chargebacks now being processed are not cases of a stranger using someone else's card.

That distinction matters, because the two explanations point to completely different fixes. If the problem is dishonest consumers, the answer is better fraud detection. If the problem is channel design, the answer is largely in the merchant's own hands — and considerably cheaper than the third-party solutions currently being sold as the solution. It was also the question we were asked directly on the panel: what can retailers actually do about it? We'll get to the practical answer, drawn from PDI's capability audits with merchants, further down.

Most consumers don't know the word "chargeback." They know "refund," and more specifically, they know the feeling of something's wrong and I want my money back. They don't distinguish between a merchant-issued refund and an issuer-forced chargeback — nor should they be expected to. From a jobs-to-be-done perspective, the job is singular and simple. The mechanism a consumer reaches for is whichever one is easiest to find and most trusted to work.

This is where channel familiarity and accessibility do a lot of quiet work. A consumer interacts with their banking app daily. They know exactly where the "something went wrong" button lives. A merchant site, by contrast, might be visited once. If that merchant has no obvious contact path — no visible email, no chat function, a buried returns policy — the banking app isn't just easier, it's the only path the consumer can actually find.

This is the part of the conversation merchants tend to underweight: a merchant-facilitated refund and a bank-initiated chargeback are not two versions of the same cost. A returned product processed directly with the merchant typically involves the cost of the return itself and some internal processing time. A chargeback brings scheme fees, gateway mark-ups, and — per Javelin Strategy & Research's 2025 survey of U.S. issuers and merchants (commissioned by Mastercard) — an average combined cost of $46 in third-party fees and $82 in internal handling per case for merchants, frequently exceeding the value of the disputed item itself.

Put simply: every dispute that gets deflected into a chargeback instead of resolved directly with the merchant is a cost multiplication event, not a neutral rerouting. And the data suggests this deflection is common. In the same study, cardholders bypassed the merchant entirely in 75% of disputes — despite merchants successfully resolving the issue in 75% of the cases where a customer did reach out first. That gap represents a significant pool of disputes that were, in principle, solvable at the cheapest possible point in the chain, and weren't.

One thread from the panel worth sitting with: raising a chargeback can become a learned reflex, not a one-off decision. A consumer who has one smooth, successful chargeback experience with their bank is more likely to reach for that mechanism again — and increasingly, to share the tactic. Tips on how to word a dispute, which reason codes get fast resolutions, and which merchants "always cave" circulate openly on social platforms. None of this requires bad faith on the part of any individual consumer in isolation; it's the aggregate effect of a low-friction, high-success channel being publicly reverse-engineered.

The Javelin research offers a useful proxy for how normalised this has become: 77% of merchants and 83% of issuers surveyed agree that it has simply become too easy for consumers to dispute a transaction. Both sides of the transaction are converging on the same read of the problem — which suggests the issue sits upstream of any single institution's controls.

Not every friendly fraud case originates with the consumer. Several categories of dispute are, on inspection, self-inflicted:

Promotional terms that disappear. A "free gift with purchase" or bonus loyalty points offer that isn't honoured — whether through genuine fulfilment failure or a consumer misreading the terms — routinely ends in a dispute rather than a support ticket, especially where the promotional terms themselves are no longer visible on the merchant's site once the campaign ends. If the evidence a merchant needs to defend a dispute isn't published anywhere by the time the dispute is lodged, that's a self-created evidentiary gap.

Subscription and free-trial mechanics. Cancellation flows that are harder to complete than the original sign-up are a well-documented driver of dispute volume. Javelin's research points to the same pattern from the issuer side: the growth in digital subscription services has coincided with more customers using the dispute function rather than the cancellation button, whether because they'd genuinely forgotten the subscription existed or because disputing was simply less effort. This isn't only a card problem, either — subscription-related disputes are also recorded through BECS, and solutions are starting to emerge outside the card rails too, whether through scheme-led initiatives or smart contract mechanics being explored on PayTo. Regulation and a consistent consumer experience across these payment methods are still lagging behind the mechanics, though, so the underlying dispute risk hasn't gone away — it's just distributed across more rails.

Product depiction versus delivered reality. Unclear delivery windows, thin product descriptions, and — increasingly — overedited or AI-generated product imagery widen the gap between what a consumer expected and what arrived. That gap is where "not as described" disputes live.

Refund policies that undersell consumer rights. Where a merchant's stated refund policy is narrower than what consumer law actually provides, consumers who sense (correctly or not) that they're being short-changed will divert to the channel where they trust their rights will be enforced — their bank.

None of these are fraud-prevention problems. They're basic operational hygiene problems, and they're comparatively cheap to fix relative to a chargeback automation platform.

Unrecognised-transaction disputes have been in decline for several years, largely because issuers have invested heavily in enriching transaction descriptors so a charge on a statement more reliably matches a name the cardholder recognises. Our working hypothesis is that this trend reverses as commerce becomes increasingly agent-mediated, for a few compounding reasons.

First, an agent acting on a single instruction can generate multiple, separate charges — a multi-basket purchase split across merchants or suppliers to fulfil one intent — and those charges won't necessarily appear related to one another on a statement. Second, deferred payment execution stretches the gap between the moment a consumer expresses intent and the moment a charge actually lands, which further strains recognition; the further the charge sits from the original ask, the harder it is for a consumer to place it. Both dynamics point toward more disputes, not fewer, unless statement presentation evolves alongside the purchasing model.

In an ideal state, agentic transactions wouldn't appear as isolated line items the way conventional transactions do. They'd need better structure and grouping to show their relationship to the original purchasing intent — for example, surfacing that a cluster of charges all belong to a holiday an agent researched and booked on the consumer's behalf, rather than presenting them as unconnected entries a cardholder has to piece together themselves. That's a theme we'll be returning to as agentic commerce protocols mature; it's not yet a fully solved problem for any part of the ecosystem.

This was the direct question put to the panel: what can retailers actually do about their chargeback numbers, in practical terms? Through PDI's capability audits with merchants, we've landed on eight areas worth working through in order — deliberately before any conversation about third-party fraud or chargeback automation software. The point isn't that automation software is only ever reactive; it's that fighting a chargeback which could have been prevented is a slower and more expensive path than preventing it in the first place. The most effective chargeback is the one that never gets lodged, and most of what determines whether it gets lodged sits inside the merchant's own operation.

1. Understand your true numbers. Not just chargeback volume and cost, but chargeback cost by product, category, and channel — and the margin pressure that creates against each. Without this, every subsequent fix is a guess. Which SKUs are disproportionately disputed, and does that correlate with delivery timeframes, price point, or something about how the product is presented?

2. Audit product and checkout pages. Descriptions that undersell complexity, delivery estimates that don't reflect reality, and product imagery that's been overedited or AI-generated to the point of misrepresenting what actually arrives are all direct inputs into "not as described" disputes. This is a page-by-page review, not a policy statement.

3. Review communication processes and friction holistically. This covers the full arc of customer contact — shipping and delivery updates, order status visibility, and whether there's an obvious hook (email, chat, contact form) for a customer to raise a problem before it becomes a bank-side dispute. The test is simple: can a customer find a way to talk to you in under a minute, without a phone call?

4. Build genuine self-service. A well-signposted, easy-to-use self-service flow for returns, refunds, and troubleshooting absorbs volume that would otherwise default to the banking app — the channel consumers already trust and already know how to use. This doesn't require instant resolution or an instant refund. What it requires is giving the consumer confidence that their request has actually been lodged with the merchant and will be responded to within a reasonable, communicated timeframe. Confidence that something is being handled is often enough to keep a consumer from reaching for their bank instead.

5. Align refund policy with consumer law. Where a stated policy is narrower than what consumer law actually provides, customers who sense they're being short-changed will bypass the merchant entirely. This is a compliance check as much as a customer-experience one. It's also worth noting that banks do, at times, check a merchant's terms and conditions when assessing a dispute — so making those terms easily accessible and machine-readable, rather than buried in a linked PDF, can help a case get resolved or deflected before it ever reaches chargeback lodgement.

6. Fix free gift and promotion mechanics. Ensure promotional and bonus terms are fulfilled as advertised, and — critically — keep those terms published and findable well past the promotion's end date, so they exist as evidence if a dispute does arise later.

7. A/B test first-party abuse rules before assuming they'll backfire. Velocity limits and additional reasoning steps ahead of a refund being granted are often assumed to drive customers toward chargebacks instead. In the cases we've reviewed through PDI's audits, that assumption hasn't consistently held — added friction at this specific point has not automatically translated into higher chargeback volume. There's an important distinction here, though. Consumers tend to accept it when a merchant pushes back on a refund request with a stated reason — for example, flagging that they've already lodged refund requests for a given share of their total orders, or a certain number of times in a quarter. That kind of reasoned friction doesn't appear to drive chargebacks. What does drive frustration and chargebacks is process friction — making it needlessly hard to lodge a request in the first place, with no explanation attached. The two are easy to conflate but behave very differently, which is exactly why this is worth testing on a merchant's own data rather than assuming either direction.

8. Only then, procure targeted third-party tooling for what's left. Once steps one through seven have been worked through, whatever chargeback volume remains is a much narrower, better-defined problem — and a far more targeted (and cheaper) case for automation or fraud-detection software than buying a platform to paper over gaps that could have been closed internally. Just as importantly, working through steps one to seven first makes it far clearer what kind of solution is actually needed — chargeback automation software, returns automation, chargeback underwriting, or another fraud-specific tool are not interchangeable, and they solve different problems. Merchants who go straight to a third-party solution before doing this work are, in effect, buying blind — and typically end up overspending on tooling that doesn't match the actual shape of their problem.

The industry default is to treat friendly fraud as a detection problem — get better at spotting the bad actors. The panel conversation this week pointed somewhere slightly different: for a meaningful share of these disputes, the more useful question isn't "how do we catch this consumer" but "why did our own process make the bank's dispute button the fastest path to resolution." The eight steps above are where that question gets answered in practice — not a comfortable audit for merchants to run, but the one most within their control to act on before any third-party spend.