Marketplaces have proven to be a thriving business model in the last decade, being almost always more profitable than traditional linear flow business models. A marketplace is any selling platform online that has more than one seller. Think Amazon, Uber, Alibaba.

More and more big brands are naturally choosing to make the transition to hosting a marketplace on their e-commerce website. That implies letting third party sellers register and sell their products on a website that isn’t theirs. That makes it logistical chaos, but also a fun challenge for a designer to tackle.

CatalogManager is a back-office interface that allows e-commerce platforms to manage their inventory, and for marketplace owners to manage their partnered sellers, and the products that they sell.

In the feature I’m going to showcase here, the problem is the following: As a big player on the industry, I’ve got millions of items on my marketplace. My partnered sellers enter in new products in bulk on a daily basis. And I don’t have time to sort through all of them to make sure that each item description is correct, that the pictures meet standards, or that they’ve been placed in the correct category by the sellers.

As a result, a significant amount of the items I sell on my marketplace are not in the right category. If a customer tries to shop for a desk for instance, they might see a chair lost in there that hasn’t been categorised correctly and sneaked past my radar.

This feature will help marketplace owners check new products for miscategorisation without effort and backtrack through their entire catalog in the process, correcting mistakes via machine learning.

There are two main mechanics to this feature:

• Sorting through items newly added to the marketplace by sellers to detect mistakes.
• When correcting mistakes, apply correction to older products from the catalog that might need it.

For this to be possible, we need to rely on machine learning. A machine learning program can sweep through all the existing catalog of the marketplace, and by comparing keywords in the products’ attributes (brand, description, title), it can figure out which type of product should belong in which category.

This way, if new products’ categories don’t match the existing architecture of the marketplace, the program can offer a pretty accurate correction.

When optimising the category of a newly entered product. CatalogManager can scan the entirety of the catalog, looking for similar products that might need a correction as well. If a seller registers a chair in the desk category, and I correct it. I’ll be offered to recategorise other lost chairs that are in a category other than Chair. This of course needs a lot of human input for the machine to become more accurate over time.