Open beta: Scan and Match free and unlimited until 31 January. Sign up and get founder pricing.
Methodology
The process, step by step
We track every launch
From its first day on Amazon: price, sales rank, deals and coupons, data point by data point.
AI reads every listing
Title, bullet points and description: ingredients, benefits, claims, format, images… It all becomes data.
We measure who wins
Winner = the top-selling 20% of its launch quarter, between days 60 and 180.
Sealed exam
The model is frozen and tested once only on products launched later. We publish the result whatever it is; if it fails, Match is shown as descriptive.
Your diagnosis
We compare your listing with the winners and tell you which changes bring it closest to them.
What winning means
Winner = a product in the top 20% of the sales rank of its launch quarter (median of days 60 to 180). “Times more frequent” = how many times more often the attribute appears among winners than among all products (1.5 times = 50% more). Only those that hold up when checked with products launched after the ones used to discover them are included.
Scan and Match
Scan · a snapshot of the podium. We compare today's best sellers with products that sell less and see what the top ones share: images, claims, price, certifications. It is fast, but the snapshot shows what they have, not what got them there: many were already big.
Match · the race from the start. We follow every new product from launch day for six months, whether it succeeds or not. Then we test the model with a sealed exam: the rules are fixed on one date and the exam happens on a later one, with launches the model has never seen. That is how we can tell you how close your listing is to the winners' and what it is missing.
Why both? Scan gives breadth and tells us where Match is worth doing; Match gives the proof.
Scan and Match, point by point
| Scan · exploratory | Match · exam pending | |
|---|---|---|
| Question it answers | What do today's best sellers have? | What did the products that went on to win have on launch day, and how similar is your listing? |
| Products it looks at | Today's 100-200 best sellers versus 200-400 that sell less, new and old mixed together | Thousands of launches, tracked from day one for 6 months |
| What counts as success | Being at the top today | Being in the top 20% of its cohort at 6 months (public definition in Methodology) |
| How it is checked | Only differences with a 95% interval and correction for multiple comparisons | The same, plus a sealed exam with new launches and published live hit rates |
| What you can do | See patterns, get ideas, request Match for your category | Diagnose your listing, compare two versions, assess a concept, Wintraits Match Index |
| How long it takes | Days | Weeks |
| Categories today | 11: Manual shaving (ES), Eye creams (ES), Day creams (FR), Face moisturisers (ES), Air fryers (ES), Face cleansers (ES), Hair masks (ES), Face masks (ES), Perfumes and fragrances (ES), Face serums (DE), Face serums (IT) | 2: Face serums (ES), Face moisturisers (ES) |
Wintraits Scan Index: how it is calculated
It is our summary of a listing against today's best sellers: a number from 0 to 100 and where it falls against the best sellers' band and the rest's band.
- Weight of each trait. Every listing trait we measure counts, not only the ones we publish. Each one weighs the difference in points between best sellers and the rest: if 70% of best sellers and 40% of the rest have it, it weighs about 30. Among best sellers, what the top ones have counts more: no. 1 weighs twice as much as no. 3 and more than four times as much as no. 20.
- Clear and doubtful differences. The more likely a difference is to be chance, the less it weighs: a clear one counts almost in full and a doubtful one, almost nothing. What the rest has more often than best sellers does not subtract: we never ask you to remove anything that is true of your product.
- Traits with a number. For images and title and description length, half the weight is for passing the cut and the other half depends on where your number falls among best sellers: the highest gets all of it; the median, half.
- Your points. What your listing's traits add up to, divided by the sum of the weights of those we could read, times 100.
- Your number. Your points against those of the most complete best-seller listing today, which is 100. Only the best reaches 100; the rest fall below.
- What can't be read doesn't count. If a trait can't be read from your listing (video, if you paste only the text), it neither adds nor subtracts. At least 3 traits must be read; with fewer we give no index.
- Bands, out of sample. We score every listing in the sample split into two fixed halves: each half with the weights measured in the other, so best sellers aren't inflated by their own data. The band is the middle half of each group (25% to 75%), with the same traits we read from your listing.
- The verdict. “Best sellers' zone” if you reach their band and pass the rest's, or reach their median; “rest's zone” if you stay within the rest's band, or don't pass its median; “in between” otherwise.
- Updated every month. Every month we measure each category's best sellers again and recompute the weights, the bands and the 100 with them. Every number carries the date of its data; if more than 45 days go by without an update, we say so.
It is neither a probability of sales nor a quality grade: it describes what today's best sellers have. What depends on the store (who sells it, variations, price) doesn't count, because a new brand can't change it by writing a better listing.
Where the data comes from
Public marketplace data on the products launched in each category, followed for at least 180 days. We only publish aggregated figures, each backed by 30 or more products, and never data on a specific product.
From what we count to what we analyse
- 191,247 detected: launches tracked from their first day on Amazon (all categories)
- 26,901 usable: with the listing read one by one with AI (all categories)
- 8,061 with a 180-day result (categories under analysis)
- 561 winners under this page's definition (categories under analysis)
The first two steps count every category; the last two, the 2 categories under analysis with Match. Before that, to choose categories, we counted 153,198 launches in 26 subcategories. The current model analyses 2,852 of them. From top to bottom, each step requires more data: only a launch with 180 days of history can tell whether it succeeded.
Sealed exam
Each exam is registered with a date (pre-registration) before seeing the results. The model is trained on older launches and checked only once with products launched later. The current exam is pending a rerun: The model learned from listings as they are today, not as they were on launch day: it describes what the listings of successful products look like. The accuracy figure at launch is pending a new test. We publish the result whatever it is; if it fails, Match is shown as descriptive.
We save each launch's listing exactly as it appears on day one. The exam uses only launch-day data (price, coupons, sellers, brand and date); what the listing text adds is checked later with live hits.
Each category under analysis has its own pre-registered clean exam: only with launch-day data and with launches the model has never seen. It is run once, on the registered date, and published here whatever the outcome.
We have tried 76 model variants and candidate combinations on past data, across all categories; only those chosen before the exam count.
Limits
- Correlation, not causation: winners tend to have certain attributes; that guarantees nothing for another product.
- Promises and certifications, only if you can prove them (EU Regulation 655/2013).
- We only publish a category when it meets our publication rules; the rest are still under study.
How to cite: Wintraits, [category], [data date], [page URL].