How Nova's Amazon benchmarks are built

The panel is a fixed panel of several hundred Amazon sellers whose Amazon accounts are connected to Nova. Their trading over the period covers more than 65 million units. It is a panel of real operating accounts, not a survey and not a scrape: the figures come from the same fee, order and settlement data those sellers use to run their business.

The panel is not a random sample of all Amazon sellers. It skews toward established brands and agencies that have connected their accounts to an analytics tool. Read the numbers as benchmarks from that population, not as a census of the marketplace.

For any year-on-year comparison, the panel is fixed: the same accounts appear in both windows. If accounts joining Nova were allowed to enter mid-period, growth in the aggregate would partly measure Nova's own customer growth rather than what those sellers traded. Freezing membership removes that.

The consequence is that a comparison window is smaller than the full connected base, and different periods can rest on different fixed sets. Each figure carries the period it was measured over for that reason.

Publishing totals would say something about the size of the panel, and that is commercially confidential to the sellers in it. Ratios do not. A fee expressed as a share of sales, a refund rate expressed as a share of units, or a day expressed as a multiple of a normal trading day carries the useful signal without exposing anyone's turnover.

Index values work the same way: a peak day is reported as a multiple of the trailing baseline, so the shape of the season is visible and the scale is not.

A figure is only published when enough accounts contribute to it that no single account can move it materially, and when the underlying data is complete for the whole window. Cuts that would rest on a thin set of accounts, for example a narrow category in a single marketplace, are not published at all rather than published with a caveat.

No individual seller, brand, ASIN or marketplace is identified in any figure. No absolute revenue, GMV or spend is published. No counts describing the panel are published, in the copy, in the page schema or anywhere else.

Where Nova quotes a third-party statistic, it is attributed to that publication with a link. Nova figures are never presented as third-party research, and third-party figures are never presented as Nova's.

Figures are restated only with a dated note explaining what changed and why, never silently. Each benchmark also carries an as-of date so you can tell which vintage of the data you are reading. The current set is as of 2026-09-09.

You are welcome to cite them. Name Nova first-party panel as the source, include the period the figure was measured over, and link to this page so readers can see how it was built.