How Scoutvo arrives at its conclusions — data sources, verification procedures, confidence values and the limits of the system. So you can judge every number before you trust it.
Scoutvo starts with available observations and separates them from derived or inferred results:
What we do not use: Amazon-internal data, data from other sellers’ accounts, or purchased clickstream profiles.
Reports use the observations available at the time of analysis. Check the timestamp, monitored product and event type before acting: collection intervals, delivery timing and coverage can differ. A scheduled check cannot establish what happened between observations or guarantee immediate notification.
Scoutvo uses available seller-profile and business-record signals to suggest registered seller location. Conflicting evidence is flagged for review rather than treated as proof of disguise:
A classification may show sources and a confidence indicator. Confidence is a model assessment, not a calibrated probability of correctness. Missing or conflicting evidence should remain “unclear” and be checked against the underlying records.
How we measure quality: The classification is preliminary. A usable accuracy claim requires a dated sample, sample size, class mix, independent labels, holdout procedure and errors by class. These details are not published here, so no validated accuracy percentage is promised. Seller location does not prove production country, landed costs, margin or deceptive intent.
Purchase criteria: Scoutvo extracts product features from the listings and review texts of a category and relates them to sellers’ revenue signals. Important: the result is a correlation, not proven causation — it suggests which features co-occur with stronger estimated signals in the selected pool. Review mentions describe the sampled reviewers, not all buyers; neither source proves a feature caused a sale.
Keywords: Your listing is compared against the category’s top competitors using term weighting (TF-IDF), cleaned of brand names and filler words. This surfaces candidate terms present in the selected competitor texts but missing from your listing. TF-IDF measures text weighting, not search volume, indexing, ranking, conversion or actual purchase queries. Check relevance and product-specific substantiation before using a term.
Images: Main-image recommendations draw on the evaluation of our own image A/B tests and on comparing image building blocks (badges, cut-outs, usage scenes) within the category. Image recommendations are hypotheses with a test recommendation — not a promise.
Copy suggestions may draw on review quotes, competitor text and your product data. Competitor wording is a research lead, not substantiation for your product: verify specifications, comparisons and performance claims before publishing. Variant checks identify inconsistencies in descriptions, images and specifications; review sharing depends on current Amazon policy and category eligibility.
Observed: a field visible in the collected source at its timestamp. Derived: a calculation from those fields, such as listing coverage. Estimated: a model-based quantity such as sales or revenue, not account transactions. Inferred: a tentative interpretation, such as seller location or a testing priority. Missing inputs remain unknown. Scores need their formula, input coverage and denominator to be independently reproducible; precision on screen is not proof of accuracy.
Honesty is part of the methodology. You should know these limits:
Editorial clarification: 16 October 2026. Operational coverage and schedules still require product-level confirmation. If our procedures change, we update this description.