A retailer with physical stores has two records of what customers buy. The website usually sorts by one of them. The other, often the larger one, stays in the ERP. This post is about putting it to work.
Why online sales alone can mislead
A category page is sorted by evidence, and for many products the online evidence is thin. A product that sold four units on the website last week might be a quiet seller, or it might be one of the chain's best products that happens to sit on page three, where nobody sees it.
That is the circular part of sorting by online sales: products sell where they are shown, and they are shown where they sold. A product that starts low stays low, because it never gets the views that would produce the sales that would move it up.
Store sales break the circle. In a store, the product is on a rail next to everything else. What customers pick there is evidence that does not depend on your website's current order.
Three numbers, not one
The useful structure is to keep three numbers per product and period:
- online sales, from the website,
- offline sales, from the stores,
- the total of the two.
Keeping them apart means the choice is made in the sorting rule, per collection, and not once for the whole catalog when the data is prepared.
This is how Merchandiser's product metrics API is built. Your system sends, for each SKU, the sales of the last day, week and month as online sales, offline sales or totals. When you send an online or an offline figure, the total of that period is recalculated from the two. When you send a total, it is taken as it is.
When totals are the right signal
Rank by totals when store and web customers are broadly the same people buying the same things. Core ranges, basics and anything whose appeal does not depend on the channel are good candidates.
Totals help most with:
- New products. A product that has been in stores for two weeks arrives online with evidence instead of with nothing.
- Deep catalogs. Where online sales per product are low, totals give the ranking more to work with.
- Seasonal turns. Stores often show the shift to a new season before the website does.
When to stay with online sales
Totals mislead when the channels differ.
- Store-only favorites. Some products sell in stores because they are by the till, or because customers want to try them on. Online they may not move at all.
- Different assortments. If the stores carry lines the website barely stocks, their sales say little about what to show online.
- Different customers. A chain whose stores serve one region and whose website serves the whole country has two audiences.
In these categories, rank by online sales and treat store sales as a second opinion to look at, not as part of the score. Because the numbers are kept apart, both choices can live in the same store.
How the data gets there
The sales already exist, in the ERP or in the data warehouse. What is needed is a small scheduled job that sends them:
- Collect units sold per SKU for the last day, week and month, split by channel.
- Send them to the product metrics API in batches.
- Check the answer. It reports how many products were updated and lists SKUs it did not recognize, which is usually a sign that the two systems name a product differently.
SKUs are the join. If the ERP and the website use different codes for the same product, settle that first; no ranking logic makes up for a mismatch there.
Using the numbers
Once the totals are in, they are ranking signals like any other. A rule can use them directly, for example total weekly sales scaled by size availability. Merchandiser AI learns from the totals when they are present, so the model sees the demand of both channels.
A sensible first step is to change one large category to a rule that uses totals, preview the order and compare it with the current one. The products that jump are the ones your stores knew about and your website did not. Look at a few of them. If they are products you would be glad to see higher, widen the change. If they are store-only favorites, that category is better left on online sales.
For choosing between a rule and the model for a category, see Sorting Rules vs. AI Ranking.