Why most store sales estimates are wrong, and how we model ours
Every store sales figure you see from outside is a model. Here is why those models go wrong, exactly how RivalSonar builds its estimates, where ours is weak and how to use them well.
On this page
Nobody outside a store can see its orders. Not us, not any other research tool, not the person in a forum who swears a store makes a certain amount a day. Every store sales figure you find from outside is a model built on public signals. The question is not whether it is exact, because it is not. The question is whether it is built sensibly and whether you know how far to trust it.
This post explains why sales estimates often go wrong, then shows exactly how RivalSonar builds its own, including the parts we are least sure of.
Why estimates go wrong
They hide the method
Many tools show a single confident number with no explanation. When you cannot see how a number was made, you cannot judge when it is likely to be off. A number without a method looks precise and behaves like a guess.
They count the wrong thing
Some trackers infer sales from stock levels or catalogue changes alone. That can work on stores that expose stock, but many stores do not, and some change their catalogue for reasons that have nothing to do with sales, such as restocks, price tests or seasonal edits. A model built on one signal breaks whenever that signal does.
They ignore what kind of store it is
A store selling a cheap phone accessory and a store selling furniture do not convert visitors at the same rate. Applying one conversion rate to every store makes cheap stores look smaller than they are and expensive stores look much bigger.
They treat small stores like big ones
Public traffic data is good at telling big sites apart and weak at telling small sites apart. Most stores are small. Estimates for the long tail are where the largest errors live.
How RivalSonar models a store's sales
Our model has four steps. Each one is simple on purpose, so you can follow it and check it.
Step 1, estimated visits
We start from public traffic rankings. Sites are ranked by popularity in public lists, and we convert each rank to an estimated number of monthly visits with a rank-to-visits curve. Visits fall steeply as rank goes down, then the curve flattens across the long tail of small sites. This gives every store an estimated monthly traffic figure.
Step 2, estimated orders
Estimated orders are estimated visits times a conversion rate. We start from a base of about 2%, then adjust it by the store's price band and category. Cheaper products and impulse categories convert higher, expensive products and considered purchases convert lower. The result is kept between 0.3% and 5%, so no store gets an extreme rate.
Step 3, estimated revenue
Estimated revenue is estimated orders times an estimated average order value. We set it at about 1.2 times the store's typical product price, because many orders contain more than one item or an add-on. This gives a figure closer to GMV than to profit, since it does not subtract refunds, fees, product costs or ad spend.
Step 4, orders per product
A store's estimated orders are split across its best sellers with a power curve. The product at the top of the best-selling order takes the biggest share, the second takes less, and so on down the list. That is how each product gets its own estimated orders.
Hourly refinement for tracked stores
When you track a store, we check its public catalogue every hour. Changes in the catalogue are used to refine the store's estimated orders and to show which products sold. This is why tracked stores usually have better estimates than untracked ones. See our post introducing the sales tracker.
Where our model is weak
We would rather you know this than find out the hard way.
- Small stores. A small change in public rank can mean a large change in estimated visits, so small-store estimates swing more.
- Unusual conversion. Stores with strong brands, loyal repeat buyers or cash on delivery offers can convert far above our range, and stores buying cheap traffic can convert far below it.
- Unusual order values. Stores that sell mostly bundles, or one very expensive item, can differ a lot from the 1.2 times rule.
- Off-site sales. Orders taken through marketplaces, social shops, wholesale or phone are not on the storefront and are not in our estimate.
- Catalogue noise. For tracked stores, restocks, price edits and catalogue clean-ups can look like sales activity.
How to use estimates well
Compare, do not quote
Estimates are most useful side by side. If two stores in the same niche show estimated orders of a similar size, they are probably of a similar size. If one shows ten times the other, it is probably much bigger. The exact figures matter less than the gap.
Watch trends
A store whose estimated visits and orders rise for three months in a row is growing, whatever the precise level. Direction over time is more reliable than any single month.
Think in ranges
Read an estimate as the middle of a range, not a point. For a decision that matters, ask what you would do if the real number were half the estimate, and what you would do if it were double. If the answer is the same, the estimate is good enough.
Check against other signals
Long-running ads, a high best-seller rank for the same product across several stores and steady growth in the catalogue all support an estimate. If an estimate says a store is large but it has no ads, little traffic from search and a thin catalogue, be suspicious.
Read more
The guide on how to estimate any store's sales shows how to run this method by hand, with a worked example, so you can check any store without us.
