Data and methodology

Client facing document, handed over on request. Version of September 30, 2026. Every figure in this document carries its date and its scope.

1. Where the data comes from

WhoHeld works from what the Amazon product page shows to any visitor: the price, the seller who holds the Buy Box, the offers listed, the sales rank, the review count. We obtain these series from a market data provider established in Germany, under a standard paid plan and a written commercial license. The provider has confirmed in writing that our use is not a resale of its data. The name of the provider and the attestation are given to your procurement and data protection teams on request, under a confidentiality agreement.

Nothing comes from your vendor account. We never ask for Vendor Central or Seller Central access, API keys, or any connection to your systems. We process no buyer data, no order data, no payment data.

2. How the data is collected

Collection is external and shared: a product is collected once, and the result feeds every client scope that declares it. Two levels of tracking exist.

Your catalog. The products you declare are refreshed by activity: every 24 hours for the products on your Hyper Care list and for 30 days after a Buy Box loss or suppression, at least every 28 days otherwise. In Lead+, a hero pack of 50 products of your choice is checked every hour. Each refresh brings back the full Buy Box history since the previous one, with a resolution of one to six hours, depending on how popular the product is.
The market. The wider category around your products is observed monthly. It measures the head of each shelf, not intraday moves. It is outside the freshness commitment, and the contract says so.

The contract commits to a minimum number of observations per month per product, set in the order form. The observed cadence is measured and shown in the application. Every screen shows when its data was last refreshed. We never say "real time", because nobody has it honestly.

3. How we rebuild the Buy Box history

The raw series are event based: a new point appears when something changes. Counting points instead of measuring time gives wrong answers. When we first counted suppressed Buy Box events, we found 14.3 percent. When we measured time instead, the real figure on the same products was 52.4 percent. We published the correction (July 2026), and every indicator has been measured in time since.

For each product and each market, we rebuild who held the Buy Box, from when to when. Holders are grouped into camps, as you declare them: yours (Amazon Retail and the partners you appoint), other sellers, and suppressed (no Buy Box at all). When the source does not know who held the Buy Box, we show "unknown"; when the listing carries no offer at all, we show "no offer on the listing". Both are left out of every share and shown apart. We never count an unknown holder as yours.

4. What is a fact, and what is an estimate

Facts, measured and dated: hours held by each camp, the number of switches, the presence of a seller on your product, the Buy Box price, the periods with no Buy Box at all, and the monthly sales volume when Amazon publishes it.

Estimates, always shown as a range and never as a unit count: the sales behind a rank, the exposure in money, the weight of demand. A product with no published sales signal is treated as "under 50 units per month", and we show how much of your catalog is covered by a direct signal, both as a count of products and as an upper bound in money for what is not.

Demand weighting. One hour of Buy Box on a product that sells ten thousand units a month is not one hour on a product that sells twenty. Classic tools count them the same. We weight Buy Box share by real demand, and we show both readings. On the brands we measured, the difference is 20 to 35 points per country. The full method is handed over on day one.

5. What we never do with the data

We never attribute a cause we did not observe. When the Buy Box is suppressed, we say what we saw: no eligible new offer, or a lowest price above the competitive threshold Amazon shows, or no offer data at all. We never write "because of seller X".

We never suggest a price, a target price, or a price move, for you or for a reseller. The contract forbids the use of the service as a resale price policing tool. We never qualify a seller as authorized or not: we separate the sellers you declared as partners from the ones you did not.

We show money by product, as a range, never next to a named seller.

6. The limits we publish ourselves

The catalog cadence depends on each product's activity (section 2). Between two refreshes, a screen shows the last measure, with its date. The measured cadence is shown in the application.
The market scope is a sample, refreshed monthly, with no freshness commitment. Your own products follow the catalog cadence; the wider market does not.
The cause of a Buy Box loss is univocal on a minority of hours. "It is not the price" is often our strongest conclusion.
Hours where Amazon held the Buy Box but the page showed a shipping delay of one month or more are counted separately (correction of September 2026).
Every method correction is dated and reported to you. A figure that goes down after a correction is a sign the method works, not a sign that something is hidden.

7. Retention of seller data

The name and country of a seller are kept 24 months after the last observation; the seller marketplace identifier is kept 36 months, rolling. Buy Box events and price history linked to a seller are kept 36 months. Seller profile indicators are kept 12 months. Series linked only to a product, with no seller identifier, are not personal data. Observed sellers have a public information notice, in five languages, and a free right to object, applied at every collection.