Chargeback Research: Why Customers Dispute Instead of Contacting You

Answer first: under the Visa Core Rules, a cardholder filing a consumer dispute must certify that they attempted to resolve it with the merchant. So every dispute in the 13 series is, on its face, a sworn statement that your own resolution path failed. Chargeback teams answer these with evidence and win or lose the money. Almost nobody interviews the customer to find out why the merchant path was skipped or abandoned, which is the only question whose answer reduces next quarter's dispute rate. Platforms like Koji make that interview practical, because the population is small, scattered, and unlikely to accept a call from the company they just disputed.

A dispute is a research artefact, not just a loss

Most organisations treat chargebacks as a payments problem. It gets a budget line, a fraud tool, a representment workflow, and a win rate. All of that is downstream of a decision the customer made earlier, and that decision is where the recoverable value is.

The Visa Core Rules and Visa Product and Service Rules, in the edition effective 18 April 2026, organise disputes into categories. The 13 series is titled Consumer Disputes, and its conditions are worth reading as a list of product failures rather than a list of loss codes:

Look at the right-hand column. Not one of those teams routinely receives dispute data. The reason code is a diagnostic label that the payments industry has already computed and attached to each event, at no cost to you, and in most companies it terminates in a finance dashboard. The card networks are running a free, structured, continuously updated triage of your customer experience failures, and the results are being filed as a loss category.

That framing also explains why dispute-rate reduction programmes so often stall. A representment workflow optimises the win rate on disputes that already happened. It has no mechanism for reducing the number that happen, because winning a 13.3 dispute establishes that the cardholder authorised the transaction. It establishes nothing about whether the product matched the description.

The certification that makes this a support metric

The rules do not merely suggest that customers try the merchant first. For Dispute Condition 13.3, Not as Described or Defective Merchandise/Services, the supporting documentation requirements include certification of, among other items:

That the Cardholder attempted to resolve the dispute with the Merchant

with parallel requirements elsewhere in the 13 series, including, for services, the date the cardholder cancelled or requested a credit, and for merchandise, the date they returned it or attempted to return it. Other conditions carry the requirement explicitly in the dispute rights themselves, such as Dispute Condition 12.6, where "the Cardholder must attempt to resolve the dispute with the Merchant or the Merchant's liquidator, if applicable, before the Issuer may initiate a Dispute."

Take that seriously and the accounting changes. A consumer dispute is not simply revenue reversal. It is an attested claim that your customer tried to reach you about a problem and did not get it solved. Whether the attempt was genuine, half-hearted, or a formality is precisely the empirical question, and it is not answerable from your side of the transaction because the defining feature of these customers is that they stopped talking to you.

The asymmetry you are actually competing with

There is a persistent industry framing in which customers who dispute rather than contacting support are committing something between laziness and fraud. Industry surveys have reported that a large majority of cardholders find filing a chargeback simpler than a merchant's dispute process, and that a substantial share go to the bank first. These figures come from vendor research rather than regulators, so treat the exact numbers with care, but the direction is not seriously contested.

The direction is also not surprising, because the asymmetry is deliberate and statutory. Under Regulation Z, 12 CFR 1026.13, which implements the Fair Credit Billing Act at 15 U.S.C. 1666, a consumer who sends a written billing error notice within 60 days of the first periodic statement reflecting the error obtains a strikingly protective set of rights:

• The creditor must mail or deliver written acknowledgment within 30 days, and must comply with the resolution procedures within two complete billing cycles and in no event later than 90 days (1026.13(c)).
• The consumer need not pay the disputed amount, and the creditor may not try to collect it, while the dispute is pending (1026.13(d)(1)).
• The creditor may not make or threaten an adverse credit report because of the unpaid disputed amount (1026.13(d)(2)).
• The creditor may not accelerate the debt or close or restrict the account solely because the consumer exercised these rights in good faith, and may be subject to the forfeiture penalty under 15 U.S.C. 1666(e) for failing to comply (1026.13(d)(3)).
• The creditor must conduct a reasonable investigation and, if it finds no error, explain its reasoning and furnish documentary evidence on request (1026.13(f)).

Now compare. One written notice, sent once, within a generous window, shifts the investigative burden to a bank, suspends the obligation to pay, and forbids credit-file retaliation. Your support process, however good, asks the customer to explain the problem to someone, wait, possibly explain it again, and hope.

You are not competing with fraud. You are competing with a statutory process that Congress engineered to be low-friction and low-risk for the consumer. That reframing matters because it changes the intervention. If disputes were a moral failure, the fix would be deterrence. If disputes are a rational response to a friction gradient, the fix is to reduce the friction on your side, and the only way to know where your friction is is to ask the people who chose the other path.

What to ask, and who to ask

The population is defined by your dispute records: cardholders who filed against you in the last 60 to 90 days. It is small, it is unhappy, and it is the highest-signal group in your entire customer base. Contact them with a neutral, non-collections framing that makes clear you are not challenging the dispute or asking them to reverse it.

Three design notes.

The "if no, why not" branch is the study. Everything else is context. The answers cluster into a small number of recurring causes: could not find how to contact you, tried and got no reply, expected to be refused, did not know a refund was possible, was in a hurry, or had already been told no. Each implies a completely different fix, and they are indistinguishable in your dispute data because the reason code is the same for all of them.

Ask about elapsed time. The gap between the problem occurring and the dispute being filed is your intervention window, and most companies have never measured it. If the median is eleven days, a proactive outreach on day three is a viable programme. If it is same-day, it is not, and you should spend the money on the checkout page instead.

Report the small-sample side as coverage, not percentages. Seven of nineteen disputers said they could not find a contact route is a finding. Thirty-seven percent implies a precision that twenty interviews do not support.

Why AI interviews are close to the only workable method here

This population is uniquely hostile to conventional research. They are annoyed, they have already escalated over your head, and a phone call from the company they just disputed reads as an attempt to talk them out of it. Response rates to moderated outreach are dismal, and the people who do accept skew heavily toward those willing to have a confrontation, which is exactly the wrong sample.

An asynchronous AI interview changes the dynamics in three specific ways.

No scheduling. Nineteen people across a dozen time zones is a scheduling problem that kills studies. Koji sends a link; participants respond by voice or text whenever they choose.

No confrontation. There is nobody to argue with and nobody to persuade, which removes the reason most people decline. It also removes the reason many who accept perform their grievance rather than describing it.

More candid answers about their own conduct. Some of what you need to hear is unflattering to the respondent: they did not look very hard for the contact link, or they knew the bank route was easier and took it. A well-established body of research on self-disclosure finds people report more openly, and with less impression management, when they believe an automated system rather than a person is receiving the answer. For a study where a chunk of the truth is mildly embarrassing to admit, that is a validity requirement rather than a convenience.

And the AI interviewer probes where a survey cannot. "I could not get hold of anyone" is not actionable. Koji's automatic follow-up asks what they tried and what happened, and the answer names a channel, a wait, or a dead end you can go and look at. All six structured question types described in structured questions in AI interviews (open_ended, scale, single_choice, multiple_choice, ranking, and yes_no) sit in the same conversation, so the effort score and the story behind it come from the same person.

Connecting disputes to the rest of your research

The 13 series maps cleanly onto problems the rest of this documentation treats separately, which is the practical payoff of reading dispute codes as a research backlog.

• 13.2, cancelled recurring transactions, is the payments-side signature of the failure described in auto-renewal and cancellation research: subscribers who believe they cancelled and did not. If your 13.2 volume is meaningful, your cancellation flow is producing false completion beliefs, and the dispute is the receipt.
• 13.3, not as described or defective, sits directly against warranty comprehension research. A customer who believed the warranty entitled them to a refund, and was offered a repair, has a remedy expectation gap that resolves into a dispute.
• 13.5, misrepresentation, is an advertising-claims problem, and the discipline for preventing it is the one in AI claims substantiation and health and nutrition claims: test what the claim causes people to believe before you run it.
• 13.6, credit not processed, is the enforcement end of money-back and satisfaction guarantees, where 16 CFR 239.5 conditions the advertising claim on prompt and full performance.

None of that mapping requires new instrumentation. It requires sending the reason-code breakdown to four teams that are not currently on the distribution list, and running one interview study to find out which of the plausible causes is the real one.

What good looks like

• A named cause for every reason code above 5 percent of dispute volume, sourced from customer interviews rather than inferred internally.
• A measured intervention window, so proactive outreach is designed against real elapsed time.
• A contact-attempt rate, comparing what customers say they did against the certification they signed. A large gap in either direction is informative: it means either your support records are missing contacts, or the certification is being made loosely.
• An effort differential under two points on your scale item between dealing with you and dealing with the bank. Anything larger and the friction gradient will keep routing customers away from you regardless of policy.
• Fixes shipped against the top two causes before the next quarterly review, with dispute rate by reason code as the read-out.

The honest limit

This research will not tell you which disputes were illegitimate, and it should not be used to try. Respondents who intended to misuse the process are unlikely to say so, and building a fraud-detection programme on self-report would be poor method and worse ethics. Fraud and first-party misuse controls are a separate discipline with separate tooling.

What this research does is answer the question your representment workflow structurally cannot: among the customers who had a real problem, why did your resolution path lose to the bank's. That question has an answer, the answer is actionable, and the people who hold it are already identified by name in a system you own.

Frequently asked questions

Do card network rules really require customers to contact the merchant before filing a chargeback?

For consumer disputes, the rules require the cardholder to attest to it. In the Visa Core Rules and Visa Product and Service Rules edition effective 18 April 2026, Dispute Condition 13.3 (Not as Described or Defective Merchandise/Services) lists among its supporting documentation a certification "That the Cardholder attempted to resolve the dispute with the Merchant," alongside dates for cancellation, credit request, or return. Some other conditions state it directly in the dispute rights, such as Dispute Condition 12.6, where the cardholder must attempt resolution before the issuer may initiate a dispute. In practice the strength of that attempt varies enormously, which is exactly why interviewing disputers is informative.

What is the difference between a chargeback and a Regulation Z billing error dispute?

They are different mechanisms that often describe the same customer action. A chargeback is a card network process governed by the network rules between issuer, acquirer, and merchant. A billing error dispute is a statutory consumer right under the Fair Credit Billing Act, 15 U.S.C. 1666, implemented by Regulation Z at 12 CFR 1026.13, which runs between the consumer and their creditor. Regulation Z gives the consumer a 60-day window to send a written notice, requires acknowledgment within 30 days and resolution within two billing cycles or 90 days, and prohibits collection, adverse credit reporting, and account restriction while the dispute is pending. As a merchant you experience the network process, but the consumer incentives are shaped by the statutory one.

Why interview customers who disputed instead of just improving fraud detection?

Because fraud tooling and dispute research answer different questions. Fraud controls address whether a transaction was authorised. They say nothing about why a customer who genuinely had a problem chose their bank over your support desk, which is the cause of most consumer-dispute volume in the 13 series. Winning a representment establishes that the cardholder made the purchase; it leaves the underlying product, fulfilment, or refund failure completely intact, so the same dispute recurs next month with a different customer.

How many disputers do I need to interview?

Fewer than you would expect, because you are after mechanism rather than prevalence. Ten to thirty interviews will typically surface every recurring reason customers skipped your support path, since those reasons cluster into a small set. Report the results as coverage rather than percentages: "seven of nineteen said they could not find a contact route" is defensible, whereas "37 percent" implies precision the sample cannot support. If you want a reliable rate on a specific belief, that is a separate, larger study.

Will disputing customers actually talk to us?

More than you expect, if the framing is neutral and the format is asynchronous. The failure mode is a phone call from the company they just disputed, which reads as an attempt to reverse the dispute and produces both refusals and a combative sample. An AI-moderated interview they can complete by voice or text at their own convenience, with an explicit statement that you are not challenging the dispute, removes both the scheduling barrier and the confrontation. It also tends to produce more candid answers about the respondent's own conduct, which is a meaningful share of what you need to learn.

What should I do with dispute reason codes once I have them?

Route them to the teams that can act on the cause rather than filing them under losses. 13.1 belongs with fulfilment, 13.2 with whoever owns the cancellation flow, 13.3 with product and merchandising, 13.5 with marketing and claims substantiation, and 13.6 with refund operations. The card networks are already performing a structured triage of your customer experience failures at no cost to you; in most companies that triage stops at a finance dashboard. Pair the routing with one interview study to determine which plausible cause behind each code is the real one.

---

Ready to interview your disputers? Sign up for Koji and get 10 free credits to run your first study. Import the customers behind last quarter's disputes, ask what happened in their own words, and get a ranked list of fixes from the people who left.

Related Resources

• Structured Questions in AI Interviews - the six question types behind the study design above
• Warranty Comprehension Research - the remedy expectations that turn into 13.3 disputes
• Money-Back and Satisfaction Guarantees - the promise behind 13.6 credit not processed
• Auto-Renewal and Cancellation Research - the cancellation beliefs behind 13.2
• Product Recall Notice Research - another post-purchase message that has to reach people
• Support Ticket Analysis - mining the contacts that did reach you
• Post-Purchase Survey Guide - catching problems before they reach an issuer