Retail and eCommerce Client Story

98% model accuracy through annotation

We built a better algorithm to help categorize products

30,000 items

labeled per month to create an algorithm

98% model


The challenge

Our client is a Three-sided online eCommerce marketplace (seller, buyer, company). As sellers were posting their products, each individual/store would have the ability to create whatever titles, descriptions, and place them into any category they wanted to. This caused a poor UX and customer experience. Imagine if you searched for your favorite baseball team and a bunch of basketball and football memorabilia came up.

Our solution

We built a better algorithm to help categorize products. This provided a better user experience for buyers and created less work for the sellers. We labelled over 30,000 items per month with 6 Annotators in order to create an algorithm that allowed machine learning categorization based on the image (computer vision) and description that the seller puts in.

The results

Now when buyers search for items or categories, the results that come back are 98% accurate. This algorithm no longer needs annotation to build the model and is validated twice per year.


Contact Us