Data precision and recall on ecommerce site

By Pavel Zaslavsky, Sep 17,2020

Unlike the enterprise sellers, who can miss a tag but still make a sell, sellers that sell on a marketplaces have only one shot. If they don’t take care of their recall and precision rates they will lose sale after sale to their competition. 
Product data tagging is a big and necessary step in building a successful online store, but it’s only the first step. Once you finish setting up the navigation infrastructure and tagging your products, it’s time to check how well you did. But how do you check that?


There are two ways to assess the quality of tagging:
  • Precision measures how much of the tagging was actually correct, or how many, of all the tags you added, describe the products accurately.
  • Recall measures how many tags were correctly added to products, out of all available and relevant tags. In other words, what is the tag coverage level throughout the store.


It is always easier to understand by example, so let’s see an example of how we use precision and recall to make sure we did a good job in tagging our products. Below is an example of a product and a table that describes the attribute and value tags added to it.



In the table we see that only 4 of the total 6 tags that were added to the product describe the shoe correctly - Style, Material, Gender and Size. The Toe Style and Heel Type tags are wrong because they don't describe the product accurately (these are obviously not stilettos, and the toe is square, not round). In this case, the precision rate is 66%, because 4 out of the 6 tags we added are correct.


What does this number actually mean? If your store’s tagging precision rate is 66%, it means when shoppers will click on a certain value on the navigational menu, they will get back the relevant results only two thirds of the time. That’s far from ideal and will certainly annoy shoppers. And the other 33%? That means that in one out of three times, shoppers that looked for stiletto shoes will get these men’s shoes as a part of the results.


As for the recall, we see that the seller forgot to add the tag that indicates the color of the shoes (black). In this case, the coverage rate is at 57%, because only 4 of 7 available and relevant tags are accurately added to the product.


When your coverage rate is at 57%, that simply means that this product will come up only in 57% of the faceted searches. The rest of the time, the product will not appear at all in the results, despite the fact that it is exactly what the shopper wanted. What if this number was 40% or 30%? Can you imagine having half of your relevant products disappear from the results, because the tagging was done poorly? This may sound far fetched, but this is what happens to the majority of the sellers online.


The scenarios we described above can mean a serious blow to your online sales. If precision rates are low, and shoppers get back bad results (like men’s shoes when looking for stilettos), they will get frustrated, lose trust in the site and leave your store within seconds. If recall rates are low and users can’t find what they were looking for? They will leave the site immediately and will not come back. And why would they?


As we see, precision and recall in product data tagging have a big effect on the customer experience and sales, so if having low precision or recall becomes your concern by now, you are absolutely right. Keeping your precision and recall rates as high as possible translates into better product exposure and more accurate results for the shoppers. This improves the overall customer satisfaction, which in turn leads to better conversion to sale.


Precision and recall effects have a slightly different impact on different types of sellers. There are two main types of sellers in ecommerce - enterprise sellers and marketplace sellers. Some sellers will be of both types:
  • Enterprise sellers are sellers that set up their own website, like Walmart or Home Depot, and ideally invested the time and efforts to establish navigation infrastructure and to tag their products.
  • Marketplace sellers set up shop in an existing marketplace, like Amazon or Ebay, that provides them and hundreds of thousands of other sellers with the content infrastructure that they usually cannot change. Marketplace sellers are only responsible for their product tagging.


When the recall rate is low, some products won’t come up in results and both seller types will lose sales opportunities. But for enterprise sellers, the customers don’t know they didn’t get all the relevant results, it doesn't really impact their user experience. For example, a buyer will choose a pair of shoes from a list of 3 items instead of 11. This, however, is only true if the customer did get back some relevant results, because if they didn’t get any results, they may leave and not return to your store.


Marketplace sellers are hit much harder than the enterprise sellers by low precision and recall rates, because they compete with hundreds of other sellers, selling the same products on the same platform. If the precision and recall of the navigation data on the marketplace products are low, customers won’t get back relevant results, or any results at all from your store, but they will still get hundreds of other relevant results. These results will be from sellers that made sure their tagging has a high recall rate.


Unlike with enterprise sellers, who can miss a tag but still make a sell, sellers that sell on a marketplace have only one shot on each new customer. If they don’t take care of their recall and precision rates they will lose sale after sale.


With tagging as important as it is for the success of online stores, sellers must always keep making sure their products are tagged well. Checking the precision and recall rates is the best way to monitor the quality of tagging in the site.

Request a demo



    PhonePlease, indicate the full number. Ex.: +9720501234567