The Chargeback Arrived. Now You Have Days, Not Weeks

A chargeback is not a complaint. It is a bank telling you it has already taken the money back, and inviting you to explain why it should not have. The funds left your balance the day the notice arrived, along with a fee you do not get back whatever happens next. That framing matters, because merchants who treat the notice as the opening of a negotiation lose on the calendar before they lose on the merits.

The interesting question now is not whether software can help. It obviously can, and the parts it helps with are specific and boring. The question is which parts of a dispute response are lookup work and which parts are judgement, because getting that boundary wrong is how shops end up paying for a tool that fills in fields nobody reads while losing every case that could have been won.

What is actually being decided?

Whether the cardholder's bank believes their customer or believes you. Nobody else has a vote. Stripe is unusually plain about its own position here, stating in its documentation that it facilitates your case but has no influence over an outcome that sits at the sole discretion of the account owner's bank. The same is true of every other processor, whatever the marketing says about win rates.

Underneath sits consumer law rather than card network preference. In the United States, Regulation Z gives a cardholder the right to dispute a billing error, and the rule sets the notice window at 60 days after the first statement showing the charge. It lists what counts, including a charge the consumer did not authorise and goods not delivered as agreed. The bank must acknowledge within 30 days and resolve within two billing cycles, capped at 90. Card network rules layer on top of that floor, which is why a cardholder often gets 120 days rather than 60.

So the dispute is a small evidentiary proceeding with a deadline, decided by someone who will spend a few minutes on it. Everything below follows from that.

What can a machine genuinely assemble?

More than most merchants realise, because the modern evidence packet is mostly structured data rather than argument. The categories are stable across processors: who the customer is, where the request came from, what was sold, what was delivered, and what was said afterwards. Each of those is a database lookup or a file attachment.

Adyen's dispute API is a useful illustration of how mechanical this has become. Merchants call an endpoint to retrieve the applicable defense reasons and the document types each one requires, then upload files inside format limits the scheme sets. It also refuses material containing passports, tax records or card numbers, which is a reminder that more evidence is not automatically better evidence.

Where each piece of evidence comes from

Read down the automatable column and the pattern is clear. The fields a machine fills perfectly are the fields your systems either recorded or did not. Software cannot retrieve an IP address you never stored. This is why dispute tooling disappoints shops that buy it after the problem appears: it is a retrieval layer, and retrieval needs something to retrieve.

Why does Compelling Evidence 3.0 change the maths?

Because it converted an argument into a checklist. Visa's CE3.0 applies to one specific condition, network reason code 10.4, which covers fraud claims in a card absent environment, the category the industry calls friendly fraud. That is the category where a real customer buys something, receives it, and later tells their bank they never authorised the charge. It is the hardest kind to answer with prose and the easiest to answer with records.

The qualifying test is precise. Stripe's implementation notes spell it out: you need at least two previous undisputed transactions on the same payment method, dated between 120 and 364 days before the disputed one. Both prior charges and the disputed charge must share either two main evidence elements, or one main plus one secondary. The main elements are the purchase IP address and the device fingerprint or device ID. The secondary elements are shipping address, customer email address and customer account ID. Device fingerprint plus device ID does not count as a pair, since both describe the same thing.

Notice what that list contains and what it does not. It contains no narrative, no screenshots, no explanation of your returns policy. It is five fields and a date range. A shop that logs those fields at checkout qualifies automatically for the disputes it is going to receive; a shop that does not cannot qualify no matter how well it writes.

What still needs a person?

Three things, and none of them are the part people expect.

The first is matching the story to the reason code. A dispute arrives with a category, and an answer that ignores the category loses even when the facts are on your side. If the claim is that the goods never arrived, delivery evidence wins and a long account of how careful your quality control is does not. If the claim is that the charge was unauthorised, the CE3.0 fields win and the tracking number is close to irrelevant. Software will happily attach everything you have, which reads to a reviewer as an inability to say which fact is the point.

The second is knowing what the customer was actually promised. Your product page is the contract, and it is a document you edit. If a dispute concerns an order from March and you rewrote the shipping estimate in June, the page you screenshot today is not evidence of anything. Very few small shops keep dated versions of their own pages, and it is the single cheapest record to start keeping.

The third is the decision to respond at all, which is where the money is.

Should you fight this one?

Often not, and the arithmetic is simple enough to write on a card. You have already paid the dispute received fee and it does not come back. If you counter, most processors add a second fee that is returned only if you win. So the wager is the countered fee against the order value multiplied by your realistic win rate for that category.

An unauthorised charge dispute where you hold two matching prior transactions is close to a free bet, because you qualify for a test with published rules. An item not received dispute where you shipped without tracking is close to a certain loss, because the one document that answers the claim does not exist. Between those poles, a rule of thumb that holds up: if the order value is smaller than the fees you would spend contesting it, accept and move on.

Stripe publishes a related number for the stage before a dispute, and it is worth knowing. An early fraud warning is an issuer report flagging a payment as suspicious, and its documentation states that 80 percent of these convert into a fraud dispute if you do nothing, outside liability shift. It also warns against reflexively refunding all of them, since many would never have escalated, and puts the sensible cutoff near your dispute fee. That is the same trade as the one above, made a few weeks earlier and more cheaply.

The work that pays happens months earlier

Everything above points the same direction. Dispute outcomes are mostly decided by what your checkout wrote down, long before anyone complained. Four records do most of the work, and none of them require buying anything.

Capture the purchase IP and a device fingerprint on the payment page. These are the two main evidence elements in the Visa test, and they are the two most commonly missing. If your checkout is hosted by your processor, this is usually a setting rather than a project.

Give repeat customers a stable identity. Guest checkout is convenient and it destroys your ability to prove a relationship. An account id that persists across orders turns two unrelated payments into a documented history, which is exactly what the rule asks for.

Keep dated copies of the pages that make promises. Shipping estimates, sizing guides, subscription terms. A monthly snapshot in a folder is enough. It costs nothing and it is the only way to answer a dispute about a promise you have since reworded.

Make the descriptor recognisable. A meaningful share of disputes start with a cardholder not recognising a line on a statement. If yours reads as an unfamiliar company name, you are generating disputes that have nothing to do with your product.

The through line with the rest of the operation is worth noting. The same checkout signals that win disputes are the ones that drive automated blocking, and pushing either too hard costs you real orders. We looked at that trade in detail in a piece on what a false decline costs compared with a fraud loss, and the conclusion transfers: the goal is not the lowest chargeback rate available, it is the highest kept revenue.

Where the tooling actually earns its price

Not in writing. A well organised packet with a chronological timeline is better than a rambling one, but the reviewer is checking whether specific documents exist, not admiring the prose. The gains come from three places.

Volume, first. If you get one dispute a quarter, a subscription for handling them is a bad trade and a checklist is a good one. Somewhere around a dispute a week, the assembly time starts to justify a tool.

Deadlines, second. Missing a response window is the most expensive error available and it is entirely preventable. Anything that turns a dispute notice into a dated task with an owner has probably paid for itself.

Consistency, third. The same claim type should get the same evidence every time, so you can tell whether your approach works. Shops that improvise each response never learn anything from their own outcomes, because no two responses were comparable.

Two adjacent pieces cover neighbouring ground: our note on which parts of returns handling are worth automating makes the same distinction between retrieval and judgement, and the walkthrough of which support tickets to hand a machine first applies the same volume test. If you are building the storefront itself and want the checkout records described above to be yours rather than a vendor's, that is the case we make for a commerce site whose code and data you own outright.

What happens after you press submit?

Very little, for a long time. Once evidence goes in, the issuer takes roughly 60 to 75 days to evaluate it, and Stripe puts the full lifecycle from filing to final decision at two to three months. There is no way to speed that up other than accepting the dispute outright. The money stays out of your balance for the duration, which is the part that hurts a small shop with tight cash: a cluster of disputes in one week can lock up a fortnight of takings until spring.

The decision is final in both directions. Neither you nor the customer can appeal it, and Stripe does not support the arbitration stage that exists further up the network rules. The one exception is a late win, where an issuer credits back outside the normal cycle and the status flips from lost to won months later. It happens, it cannot be predicted, and it is not a plan.

Two categories deserve flagging because merchants discover them at the worst moment. Some disputes cannot be challenged at all under network or local rules, and are closed as lost the day they open with no evidence stage. Separately, the payment methods that are not cards run their own processes: dispute windows commonly stretch to 180 days rather than the 120 typical of cards, the provider rather than a bank decides, and the fee structure differs. If a meaningful share of your revenue arrives through a wallet or a pay later provider, its rules are a separate document you have probably never read.

Inquiries are the happier case and the one most often ignored. Some networks open a request for information before a formal chargeback, and answering it can end the matter without a dispute fee. Leaving it unanswered reads to the issuer as acceptance, and what follows is a chargeback you are unlikely to win. An inquiry that sits open for 120 days without escalating is simply closed. Treat every one as a free chance to end the case before it costs anything.

One last thing about win rates

Any number quoted to you as a win rate is close to meaningless without its denominator. A provider that only contests disputes it expects to win reports a spectacular rate on a tiny sample. A shop that contests everything reports a poor one while recovering more money in absolute terms. Neither figure tells you what you would recover.

Track your own, split by claim type, and only from the point where you started capturing the fields above. Six months of that is worth more than any benchmark, because it is measured on your customers, your products and the specific reasons people ask for their money back from you. The first surprise for most shops is that the losses cluster in one category, and that category usually has an operational fix rather than an evidentiary one.