How PeakScout estimates sales, and how we test it
Amazon publishes no sales figures. This page shows how we estimate them, what the method cannot see and how we measure our own error.
The method
Amazon does not publish sales. It does show “bought in past month” on many listings, in rounded-down brackets. We treat that as ground truth.
- Collect rank, category and price history for amazon.co.uk products.
- For products that show the badge, record the bracket. The true figure lies between the bracket and the next step up.
- Fit a curve of sales against rank for each category, on log scales.
- For a product with a badge, report the bracket. For one without, read it off the curve and add an uncertainty band of about 80%.
- Show a range and flag whether it is anchored.
How we test it
We hide the badge on products that have one, predict from the curve, and count how often the real bracket falls inside our range. That share is the hit rate, measured per category. A category with too few anchored products is marked low confidence on every product in it.
What the method cannot see
- How closely the badge tracks real sales in every category.
- How the curve behaves in small categories with few badge examples.
- How promotions distort one month’s badge.
Why we do not quote a single accuracy figure
Published accuracy figures for research tools disagree with each other, and none is an independent test on amazon.co.uk. A per-category hit rate with its method is more useful than one headline number.
Questions
How accurate are PeakScout sales estimates?
Where Amazon shows a “bought in past month” figure we report its bracket. Elsewhere we report a range from a curve fitted to those anchored products, and test the curve by hiding the badge and checking the result. Every estimate is a range for that reason.
Why ranges and not one number?
Because Amazon publishes no sales figures. A range shows the real uncertainty, and a single number hides it.
What is the biggest weakness of the method?
Products without an Amazon badge rely on a fitted curve, which is less certain than one anchored to Amazon's own figure. We mark which is which on every product.