E-Commerce

Where the decisive record sits, and why it is usually the merchant's own order table

In most e-commerce disputes the record that decides the question is the merchant's order data: the row-level table of what was sold, for how much, when, and from which landing URL. A platform's reported conversions are that platform's attribution model applied to that platform's own view of the world. The order table is a first-party record written on the merchant's own infrastructure at the moment money changed hands. Reconciling the two is the central analytical task in most of these matters, and the gap between them is usually large and usually explicable.

The second record is the marketplace's. On Amazon the seller sees what Amazon chooses to show and Amazon holds the rest. On a merchant-controlled platform such as Shopify the merchant holds essentially the complete transactional record, with the platform holding infrastructure-level data on top of it. That single difference sets the discovery strategy, because one route is a self-service export that a cooperating client can run the same afternoon and the other is a third-party subpoena whose scope will be litigated.

Counsel weighing whether to retain an ecommerce expert witness is usually choosing between those two routes without knowing which one leads anywhere. My first question is which side of that line the decisive record falls on. My second is how long it will still be there.

What the merchant's transaction record actually contains

The fields are documented, which means a request can name them instead of asking for "the sales data." Shopify's own Order resource documentation sets out attribution fields carried on every order — the merchant-side counterpart to a platform's attributed conversion:

• landing_site — the page the customer first arrived at, including its query string, which means it preserves UTM tags (campaign tracking parameters appended to a URL) and click IDs
• referring_site — the site that referred the customer
• source_name, source_identifier and source_url — the channel the order came through and the identifiers associated with it
• note_attributes — custom name/value pairs that merchants and apps use to stamp further attribution data onto an order
• client_details — browser and session information including IP address, user agent and browser dimensions

landing_site is usually the field that settles a disagreement between a platform's reported conversions and the merchant's orders, because it is a first-party record of the URL the customer actually landed on, captured at the merchant's own server and independent of any ad platform's attribution model, cookie state or consent configuration. It survives the loss of third-party cookies. It is very rarely requested.

Two practical notes for a document request. There is a richer multi-touch object, customer_journey_summary, worth naming explicitly rather than hoping it arrives with the rest. And Shopify has been migrating from its REST API to GraphQL, so a production should state which API and which version generated it — field availability differs between them, and two honest exports can disagree for that reason alone.

Marketplace-held records, and the two routes to them

Where the client is the seller, self-service export is the first route and it is fast. Business reports, advertising reports, and — for brand-registered sellers — Amazon's Brand Analytics reporting, which includes a query-level report tying search terms to impressions, clicks, cart adds and purchases at the product level. That query-level record is unusually valuable, because it is one of very few places outside a paid-search platform where a specific search term is tied to purchases of a specific product. It matters directly in listing-hijack, counterfeit and comparative-advertising disputes. The report set, the lookback windows and the eligibility rules change, so they should be confirmed against the account in evidence rather than asserted from documentation.

Where Amazon itself holds the record, a third-party subpoena is required and the scope is routinely contested. That category includes order-level buyer data, Amazon's internal decisions about which offer was featured, enforcement actions taken against a listing, and the identity of a competing seller behind an account.

One record sits awkwardly between the two. Featured Offer history — what most people still call the Buy Box — is not exposed as a first-class report. A seller can infer its own share from business reporting, but a timestamped record of which seller held the featured offer at a given moment generally has to be reconstructed from third-party monitoring services. Those services sample at intervals, observe from one vantage point, and carry no certification of any kind, which is the first thing an opposing expert will say about them.

The reporting windows on the marketplace side are the shortest in the field

Amazon's advertiser-side reporting is the most perishable of the major platforms. Amazon's own documentation repository states that a version 3 Reporting API request can cover a maximum of 31 days, and that while a request can reach up to 95 days into the past, data is only returned for the last 65 days — described there as a known issue for Sponsored Products reports in v3. The practical effect is that roughly two to three months after a campaign, the API record is gone. What remains has to be an advertiser-side or agency-side extract taken earlier.

Amazon's attribution windows are its own, and they differ from Google's and Meta's. An exhibit that combines attributed sales across those platforms without saying whose windows produced which figure is misleading on its face, and the point is easy to make on cross.

Amazon Marketing Cloud is sometimes offered as the answer to this and it is not. Amazon describes it as a clean room that "only accepts pseudonymized information," whose users "can only access aggregated, anonymous outputs." Its ad-traffic lookback window was expanded from 13 months to 25 months in an announcement dated 11 November 2025 — and an expansion does not resurrect signals that had already aged out, so it must not be described as retroactive. More importantly, the tool is architecturally incapable of answering an identification question. It will not return a row representing too few users. It can address whether exposure correlated with purchase; it cannot address whether a particular person saw a particular ad.

Four systems counting four different things, and why they never match

Checkout is where marketing evidence and transaction evidence meet, and the numbers rarely reconcile on the first pass. Analytics counts sessions. The order table counts orders. The payment processor counts settled transactions. The ad platform counts conversions it has attributed to itself. Four numbers, four definitions, four systems, and each is internally correct.

Reconciling them, and documenting in writing why they differ, is standard and highly defensible expert work. It is also the piece most often skipped, which is why so many exhibits in this field invite a discrepancy question that the proponent cannot answer.

One arithmetic point is worth stating flatly because it appears in reports constantly: platform-reported conversions cannot be summed across platforms. Each platform attributes to itself, under its own window and its own model, and the amount of double counting is not knowable from the reports themselves. A "total conversions" figure built by adding Google, Meta and an analytics tool together is not a quantity that exists.

Listing disputes, and the order of value of the available evidence

The recurring fact patterns are narrow and they repeat. A competitor edits a shared catalog listing. A seller's images, bullets or enhanced content are copied. A listing is suppressed, or variations are merged. Backend search terms are alleged to infringe a mark. Reviews are alleged to have been manipulated.

The evidence, in descending order of value: the marketplace's own change or contribution log; a contemporaneous third-party capture taken with hash values; the seller's own screenshots; and the Internet Archive. The last two are weaker than counsel usually expects. A screenshot records what one browser rendered at one moment on one machine, under whatever personalization, logged-in state, ad blocking and experiment assignment applied to that machine, and it carries no record of the server's response or of whether the content had been edited. The Internet Archive has irregular crawl frequency, historic robots exclusions, and poor coverage of dynamic and personalized pages — and a marketplace listing is personalized and often not crawled at all.

This is the practical reason to insist on account-level exports rather than counsel's screen captures. A platform-generated export is a record generated by an electronic process or system, which is the natural subject of a certification under FRE 902(13); a copy of a produced data file is the natural subject of one under FRE 902(14). Many productions need both.

What the e-commerce record does not settle

Marketplace analytics cannot establish that a listing change caused a sales change. Ranking, competitor behavior, price, inventory status, seasonality and the marketplace's own advertising all move at the same time and none of them is held constant. The data can show that a change was made on a date and that sales moved on a date. Connecting those two is a causal claim that requires the alternative explanations to be worked through and, where possible, excluded.

A third-party featured-offer tracker cannot establish continuous state. It samples; between samples it knows nothing; and it observes from one vantage point while the featured offer can vary by customer.

An export is a snapshot at export time. It cannot support a finding that nothing was deleted or edited before the export was taken, which is why the export date belongs in the report as a material fact and why a preservation demand issued early is worth more than a broad subpoena issued late.

And a certification establishes authenticity, not accuracy. The Advisory Committee said so directly: a certification under the 2017 rules "can establish only that the proffered item is authentic," and the opponent "remains free to object to admissibility of the proffered item on other grounds." An export produced under such a certification can still be hearsay, still be irrelevant, and still be wrong. Certifying that a system produced a record is not certifying that the record is true.

Frequently Asked Questions

Published 2026-08-12