Why You Need an Advanced Taxonomy to Ace Online Fashion Retail

“The challenge of search and recommending products online is increasingly being solved with data science as e-tailers compete on personalization. Taxonomy is the new fashion tech essential,” Vogue Business  

If you’ve browsed around our recommendations and search pages, you might have come across this word several times. Before we dig into why you need to know how an advanced taxonomy is mandatory for your eCommerce strategy, let’s talk about the basics.

So, what is a taxonomy?

In the general sense, taxonomy is the science of naming, describing, and classifying items into categories. In the fashion context, product taxonomy management involves categorizing products and defining product hierarchies based on a predetermined architecture. Think of this as mapping a product’s DNA. It includes every element of a garment, including physical attributes such as product type, neckline, hemline, color, pattern and aesthetic attributes like style, occasion, activity and mood. It doesn’t just end there, other business-related attributes such as season, price and brand must also be accounted for in a good taxonomy.

As a fashion tech company with an AI Styling engine that offers personalized styling to a segment of one, the first hurdle we had to cross was taxonomy.  

The challenge with fashion taxonomy is that it is almost impossible to get the same taxonomy from different retailers. What adds to the complexity is that fashion trends are constantly changing, making the taxonomy a dynamically evolving set of information.

Why is Taxonomy so important?

As Vogue phrased it, “taxonomy is the new tech essential”. For an AI system to work (well), a good taxonomy is a pre requisite. Without this language system, it is impossible to teach the machine to do anything. AI needs a well-defined classification system with clear hierarchies and as little overlap within categories as possible. To add intelligence to any fashion process, you need to ensure a good taxonomy.

The primary objective of taxonomy is to make that conversion to Add to Cart really easy. It is established that all visitors to an eCommerce site either browse or search. To make the most of that, a well-built taxonomy allows websites to display products under meaningful and well understood categories. Additionally, this allows the retailer to offer better navigation, filtering and recommendation options to make discovery easy and drive conversions.

As for quality search, a rich taxonomy captures all the many ways a shopper searches for products, understand the intent and serves fashion relevant results. This must include spelling errors, typos and vague descriptions that are bound to happen.

A Forrester research report found that poorly architected retailing sites sell 50% less than better organized sites. Where searches failed, 47% of users gave up after just one search, and only 23 percent tried three or more times.

Another critical benefit of a single, rich taxonomy is that data is understood in exactly the same way across the various departments of the organization. Reporting becomes meaningful and decision making becomes data driven and effective.

Now that you know what a good taxonomy does for fashion AI, let’s dig in to the lessons we have learnt so far while building one of the world’s most comprehensive fashion taxonomies.

Taxonomy and Streamoid’s AI styling engine: What we learnt through years of research

With our goal of styling at an individual level, needless to say, we needed a very deep taxonomy.  

Here's why: if the T-shirt had a V-neck we needed to know if it was a high V or a low V. If we were styling someone with a short neck, we’d need to recommend a low V and a high V for someone with a long neck.  

But just having a taxonomy was not enough, we needed to add these attributes to the catalog images. Manually adding these meta tags is extremely tedious and time consuming, which was then ruled out as an option. Finally, we used computer vision and built over 152 machine learning models that can add these attributes automatically with high accuracy.  

With the catalog tagged with all the relevant attributes we built our AI styling engine. It has been an arduous way up to here, but seeing the results it has given us and the value it has added to our clients has been worth it.

‍

Using our complete eCommerce solution, longtime client Forever 21 India has seen a 37.4% increase in PDP views, a 29.6% increase in Add-to-cart and 3.1X revenue per user. Even with the complexity of Indian style, thanks to our taxonomy and the ML models, our client Biba has seen a 28.5% increase in Add-to-cart within a month. Our AI recommendations have helped lingerie brand Van Heusen Intimates generate 3.5X revenue per user.

Here are some of the features powered by our universal taxonomy:

• Advanced data enrichment - Better understanding of products
• Automated cataloging content - Product titles, descriptors, style tips
• Fashion Search – Advanced text search  
• Recommendations – Outfit and matching recommendations for products
• Personalization - A 1:1 styling engine  
• Theme based product and outfit curator
• Competitor Insights and Trend analysis

Congratulations, you now know more about how to choose a tech vendor and what to look for compared to 90% of your competition. Since you’ve made it this far, here are some takeaways to think about if you are building your own taxonomy:

1. Fashion taxonomy needs constant updating with the latest terminology. It is hard to do this manually and best to use a Natura Language Processing based tools

2. The taxonomy needs to understand shopper behavior, capture all the synonyms and the many ways they search for products

3. For cataloging it is adequate to have a broad taxonomy without too much depth

4. For personalization you will need to have a broad and deep taxonomy.

5. Taxonomies need to be optimized for countries to keep it relevant

6. A lot of companies may not have any one dedicated to building and maintaining the taxonomy. If so, they should consider partnering with a tech company with expertise in the domain.