The Problem
Camicado, a leading retail brand with a significant presence both in physical stores and e-commerce, grappled with a pressing challenge regarding its online platform.
After Camicado expanded its e-commerce product catalog and integrated a third-party system to add marketplace products, some problems emerged. The systems involved didn’t have a set number of categories or a way to match products with existing categories. For example, if a marketplace product was a “wine glass,” it would create a new category called “wine glasses” in the kitchen department, instead of placing it in the existing “glasses” category under kitchen.
This caused products like wine glasses to be split into many different categories, making it harder to find them in one place. For instance, you could find different types of wine glasses in various departments, even in unrelated ones like appliances. In scale, several categories were created and spreaded through all of 50.000 types of products.
Consequences
That problem spread to several daily consequences creating a bigger impact that could only grow more over time, here are some of the impacts:
From the user’s point of view, we observed several negative trends in our data:
An increase in the bounce rate on our pages.
A decrease in conversion rates.
A significant rise in the abandonment rate on catalog pages after the project’s launch.
From the business’s point of view, we noticed that this problem also impacted our organic traffic through Google:
Fewer pages were being indexed.
Fewer keywords were being used effectively.
This created a growing gap in our online presence over time.
From the logistics point of view, we observed an increase in the time required at the Distribution Center for product inbound processing:
The confusion caused by the numerous overlapping categories made it difficult to correctly categorize products, leading to delays.
This research aimed not only to create a unified version of the categories but also to incorporate the user’s point of view into this unified version. Consequently, it helping to make the process for inbound and marketplace integrations more efficient and to enhance the user-friendly perception of categories and products.
My main question was:
How do users understand departments, categories, and subcategories, and how are these categories related?
Research Process
Discovery
Data Analysis (current impacts)
Stakeholder Interviews
Problem Statement
Exploring
Card sorting
Analysis of the Results
SEO Research
Testing
New Taxonomy Implementation
Listening
Results
Exploring
Card sorting preparation
To evaluate the size and the time frame for the card sorting I started to first test it with the business areas of Camicado. It was an online and closed card sorting. The testers would have to atribute types of products in the categories that was a good placement in their opnions. After this test we made some changes and started to recruit the users for the card sorting.
Card sorting
Two groups were assembled each day, mixing user profiles, to conduct discussions and analyze how people understand each other within a site. The dynamics lasted 10 days, with 10 groups of 7 to 8 people. There were 2 hours of card sorting with 200 cards to be organized, belonging to 3 main departments: Kitchen, Table, and Decor. It was an open card sorting system, where users could create new groups and name them as they wished. One group was assembled each day, mixing user profiles. They worked separately but could ask questions as needed.
Analysis
Cluster analysis is a multivariate technique that aims to group objects based on their characteristics. It classifies objects (e.g., respondents, products) according to what each element has in common with others in a specific group, considering a predetermined selection criterion.
average distance per group
Through this technique, we analyzed each group created by the users and developed an equation, a number that could represent the distance between types of products and categories. We called this the average distance per group. For example, if an item such as a wine glass compared to a beer glass had a coefficient of 0.55, it would indicate they likely belong to the same category but not the same subcategory.
On a scale, our table of distances counting for over 200 categories would look like this:
Looks complex, or at least busy. But if we add some color, subtitles can look a little better.
Merging the analysis with seo inputs
From the results we obtained through the cluster analysis along with the SEO research to give more depth in data, this merging helped us to ensure that terms are being used as the customers are researching. We also created, for the first time, a manual for URL and link building for the company to ensure that the pages are not just easily found through Google but also easy to navigate.
Testing
aSSESSMENT AND HELP THROUGH IMPLEMENTATION
In addition to what the taxonomy needs to look like, I also created a manual on the kind of research and process needed to create a new category to maintain the new organization. This includes documentation on which categories need to be merged or broken into more categories (like a from/to).
Results
165%
More indexed pages
53%
Decrease on Bounce rate (from Google audience)
10%
uplift of conversion rate
5 min
decreade of inbound process
81%
Increase on Menu interaction
600k
OF Growth on organic acess
Reflection and Future Prospect
This project enabled even bigger projects. During the research, we also collected direct feedback from the customers that led to a series of smaller and quick improvements that allowed us to drive even better results. The project implementation also allowed the business areas to review the process of inbound integration with the marketplace platform and create a new version of it. This is fundamental to keep the current organization updated according to the expectations of the users. This project raised more questions about how the process of purchasing new products can follow an analysis of what the customer is researching.
I learned a lot from this project, from recruiting users to conducting research, and understanding the importance of supporting methodologies and incremental data collected from other research to create a more accurate result.
