A&A INSIGHTS
"The product didn't arrive" — what to submit to a payment dispute, drawing on Stripe's evidence analysis
For cross-border ecommerce operators: use Stripe's analysis of product-not-received disputes and its Smart Disputes documentation to design an evidence packet keyed on the order ID, with a human review window before the dispute deadline, before considering AI or auto-submission.
日本語で読むTHE STARTING POINT
When a payment dispute claiming "the product did not arrive" lands on a shipped order, the outcome depends less on the wording of the rebuttal and more on whether the records that link the order to its fulfilment can be assembled before the deadline. Drawing on Stripe's public analysis of product-not-received disputes and its Smart Disputes documentation, this article proposes that the cross-border ecommerce operator design an evidence packet on top of the order ID, with a small human review window, before evaluating any AI product or auto-submission. Overseas headline numbers are treated as design cues, not as a promise about the operator's own win rate.
For product-not-received disputes, the evidence linking order to fulfilment comes before the rebuttal wording
A&A perspective
When a dispute claiming the product never arrived lands on a shipped order, the first thing an operator often wants to write is a careful explanation: point by point, matching the complainant's claim to the merchant's fulfilment record, in a polite but firm reply. But a payment dispute is not courtroom advocacy — it is a deadlined submission of evidence. The length of the rebuttal matters less than how much of the underlying fact chain — that this order was shipped, delivered, received, or used — can be assembled inside the window. And that evidence typically sits across warehouse systems, carrier dashboards, customer-support histories, and finance-side refund logs, each maintained by a different team on a different tool.
Hypothetical example
Consider a hypothetical cross-border ecommerce operator whose overseas shipments make up a meaningful share of monthly revenue. A small fraction of those orders comes back as a product-not-received dispute. The clerk opens the warehouse's shipping CSV, searches the carrier's tracking number, checks whether the support inbox contains any receipt confirmation, sees whether the refund ledger already lists a related credit, and packages the result as a PDF before the response window closes. Once the case count enters double digits per month, this cross-tool collection alone starts eating into the deadline. Practising "how to write the rebuttal" does not help; missed deadlines pile up because the retrieval is slow.
Stripe's evidence analysis distinguishes physical from information-based evidence
From the sources
In a July 21, 2026 blog post, Stripe published an analysis of the evidence submitted for product-not-received disputes and the associated win rates. For physical goods it reported the win-rate difference associated with combined delivery confirmation, a delivery map, and a signature; for digital products it reported the win-rate difference associated with submitting usage evidence, described on the page as digital activity and usage logs. The post explicitly frames these findings as correlations and does not claim that adding evidence causes the win.
A&A perspective
What A&A wants to take from this analysis is not a target win-rate number but that evidence takes different shapes for physical and for information-based products. For a physical shipment, the core proof is a visible, external attestation that delivery took place. For a digital product, or for the digital component of a physical sale — a downloadable file, a subscribed stream, a video view — whether the buyer received value can only be described through logs on the merchant's own systems or on the platform's telemetry. The word "product-not-received" is the same in both cases, but the source of the records, and the person responsible for keeping them, differ.
A&A perspective
For cross-border ecommerce this often shows up as a single order that contains both. An overseas physical shipment may include a downloadable manual PDF or a configuration file the buyer can retrieve from their purchase page. If the buyer claims the entire order did not arrive, submitting only the shipping proof leaves the digital half of the argument unaddressed. The judgement here is not about the elegance of the rebuttal wording; it is about whether, from the order ID, the merchant can name what was sold and which records were designed to persist. Overseas headline numbers are more useful as a design frame than as a target for the operator's own win rate.
Smart Disputes shows what an evidence packet must reach by the deadline
From the sources
Stripe's Smart Disputes documentation describes a behaviour in which, for eligible card disputes, Stripe extracts relevant evidence from its internal data, the merchant's transaction data, and cardholder data, tailors that evidence to the dispute reason code, and submits it automatically before the dispute deadline. The same page explicitly states that Stripe does not guarantee an outcome, that fees apply only when a dispute is won, that Smart Disputes is not a substitute for professional advice, and that the merchant remains responsible for the accuracy of the underlying data. It also describes a Dashboard setting to turn auto-submission off.
A&A perspective
The prerequisite behind this description is not "turn on auto-submission," but rather "decide, in advance, what can actually be retrieved before the deadline." Enabling auto-submission has no effect if the relevant data is not structured on the merchant's side to begin with; the packet arrives thin. The merchant-responsibility clause about data accuracy also cannot be honoured if the linkage between order ID, fulfilment events, and customer communication is not preserved. The evidence packet is a design object that predates any AI or automation decision.
A&A perspective
For the cross-border ecommerce operator, the operational implication is not a binary choice about turning Smart Disputes on. It is a self-test: given only an order ID today, can we produce a draft evidence packet in thirty minutes? If not, first build the state that makes it possible. If yes, then consider auto-submission on top of that. Reversing that order does not reduce the operator's last-minute load; it just shifts the appearance of win rate around in a way that is disconnected from the underlying record quality.
An evidence packet as four columns joined by the order ID
A&A perspective
One concrete way to make a packet easy to produce is to fix, in advance, four columns keyed on the order ID. The first column is the order record itself: purchase timestamp, item lines, shipping address, amount, currency, buyer email — the merchant's own primary record. The second column is fulfilment: warehouse dispatch timestamp, carrier, tracking number, carrier-side delivery confirmation, and where warranted a delivery map or signature. The third column is customer communication: messages before and after the dispute, buyer logins into the purchase page, any resend requests. The fourth column is refund and adjustment: refunds issued through the payment processor, adjustments through other channels, partial refunds if any.
Hypothetical example
In the hypothetical cross-border operator's case, the goal is that, from an order ID, a single PDF laying out the four columns is produced. Each column names its source tool and the ID or search condition used inside that tool. An order whose four columns fill without gaps has near-complete evidence. Where a gap appears, that is the operator's operational discontinuity, and by definition, the weak point in dispute response. Reading the four columns also shows how Stripe's finding about the different shape of evidence maps to the operator's own layout: usage logs for digital products belong in the second column, refund records in the fourth.
A&A perspective
Before adding any AI tool, the point of these four columns is to fix who owns which column. First column: order intake. Second column: warehouse and carrier. Third column: customer support. Fourth column: finance or payments administrator. A single sheet naming those owners is the first artefact. Any column without a named owner defaults to "whoever remembers" during each dispute, which shortens the effective deadline further. Once the four owners are set, AI and Smart Disputes-style automation slot into place as tools that shorten those owners' work, not as replacements for their responsibility.
Separate what can be automated from what a person must check before the deadline
A&A perspective
Even with auto-submission enabled, some checks must remain in a person's hands before the deadline. First, whether the order is already the subject of a refund or cancellation. Second, whether the buyer and the operator have already agreed on a resolution through a different channel. Third, whether the underlying issue is not actually "non-receipt" but partial fulfilment or a specification mismatch that changes the dispute reason. These cannot be answered from extracted data alone; they sit close to a business decision. What is safe to automate is the routine of extraction, formatting, and submission. Interpretation of the dispute reason and the decision of whether to contest at all should stay with a named human owner.
Hypothetical example
In the hypothetical operator's case, a weekly sample of the fully-packed disputes is reviewed by the named owner, who confirms whether each one is genuinely worth contesting. Recording the not-contest decisions as well as the contest decisions builds up an operational record that outlasts monthly fluctuations in headline win rate. That record often surfaces specific discontinuities — for example, a particular product line missing signature capture, or a particular country where tracking updates lag — which are second-column operational improvements for warehouse and carrier. Evidence design and operational improvement end up talking to each other on the same sheet.
A&A perspective
Overseas headline numbers are safer read as design cues than as a promise about the operator's own win rate. Transaction type, country, card network policy, sample period, and denominator all move the numbers. Stripe's post itself is explicit that the analysis is correlational and does not claim that adding evidence causes the win. When A&A advises a cross-border ecommerce operator on dispute response, the sequence is: get the four columns filling reliably; open a small human review window before the deadline; then set targets. Setting a target win rate before those two steps has no lever to move it with.
Four items to decide before evaluating Smart Disputes
A&A perspective
First, count separately, over the last three months, the number of product-not-received disputes lost by deadline expiry, the number closed with an uncontested refund, the number contested and won, and the number contested and lost. A single aggregate win rate hides the evidence-design problem. Second, colour-code the four evidence columns by whether the corresponding record can be retrieved from an order ID within thirty minutes. Anything that cannot be colour-coded is a record that only lives in someone's memory today.
A&A perspective
Third, split the AI or Smart Disputes decision by which layer it should affect: extraction speed, submission automation, or interpretation of the dispute reason. Blending the three into a single "AI-ify dispute response" narrative tends to erode the human review window rather than protect it. Fourth, share on a single sheet the four column owners and the named individual who has authority to stop a contest — typically the operator or one designated manager. Introducing automation without those responsibility boundaries lets headline numbers drift away from the shop floor's actual workload and slows down root-cause investigation when a slump appears.
A&A perspective
When A&A takes a dispute-response design engagement with a cross-border ecommerce operator, the four-column visualisation and the small pre-deadline human window come first. A related article, "Explaining return costs with product data: Wayfair's return-driver analysis," applies the same order-ID-column idea to a different revenue-leakage cause. Overseas headline numbers are useful to reshape evidence design; they are not targets to import.
Product-not-received disputes turn less on rebuttal craft and more on whether the records tying the order to its fulfilment can be assembled before the deadline. Stripe's evidence analysis and Smart Disputes documentation together make explicit the difference between physical and digital evidence, the preconditions for pre-deadline automatic submission, and the merchant's own responsibility for data accuracy. Before importing overseas win-rate numbers as targets, build the four order-ID columns and a small human review window; AI and Smart Disputes belong on top of that base, not in place of it.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- Analyzing the evidence that helps businesses win product-not-received disputes
Stripe · 2026-07-21
Accessed 2026-09-19 - Smart Disputes
Stripe · Publication date not stated on page
Accessed 2026-09-19
AI-assisted editorial production
A&A uses AI for research, writing, translation and editorial checks. Source facts, our analysis and hypothetical examples are labeled separately.
Editorial check: 2026-09-19