This article details the logic and layers of learning in how ShoeSize.Me delivers and improves its recommendations to your shoppers
The ShoeSize.Me database contains over 5000 brands and over a million models across 18 different international size scales; all of which are linked by the size and fit data of real shoppers. With every recommendation ShoeSize.Me gives, our algorithm is learning. But what happens when a new brand or model is entered into our database? How does our algorithm determine what the best possible size recommendation and learn over time?
Our machine learning technology process works in a 3 layered method. The core of which is the Purchase & Returns data exchanges:
- Size Chart matching
- Employee Shoe Shelf Building
- Purchase & Returns Data
1. Size Chart Matching
The ShoeSize.Me database contains an Official Size Chart for every brand in our database. We take the official size chart guidelines of each brand and compare and connect them to each other. On a basic level, this enables us to provide a recommendation that follows the official size guides that each of the brands uses and publishes. ShoeSize.Me is able to compare these charts against each other and calculate a recommendation. This acts as the first layer in how our algorithm learns and is a starting point when our database has little or no previous data.
2. Employee Shoe Shelf Building
We understand that sometimes having a certain level of control over the technology and the recommendations given by the Size Advisor is desired - especially in cases where there is limited or no previous data in ShoeSize.Me's database. For this, we have created ways for our partners to input directly the size and fit experiences they have of their products directly into our database. This creates direct links from their brand to other brands in the ShoeSize.Me database. Doing this provides additional data points to speed up the learning of our algorithm and will serve as another layer for our algorithm to learn from. With time, the data input may change as the algorithm learns and engages with more and more of your shoppers. The ShoeSize.Me onboarding team will guide you in how to do this.
3. Purchase & Returns Data
While the Size Chart Matching and the Employee Shoe Shelf layers described above provide a basis for recommendations, the core long-term machine learning by our algorithm is done through the daily Purchase & Returns exchanges set up during onboarding. From the Purchase & Returns, our algorithm analyzes combinations of user input, purchase patterns and return patterns and combines those with product attributes that enable ShoeSize.Me to calculate size recommendations. Through the usage of the Size Advisor by your shoppers, with time, the algorithm learns more and more about sizing differences between your shoe models and their connections to other brands and retailers. The long-term learning curve achieved through the usage of the Size Advisor and the Purchase & Returns data follows an s-curve with a slow start, steep learning, and a sustainable plateau. There is no way to “skip” or fast forward this learning, but remember; ShoeSize.Me is constantly learning from the insecure and uncertain shoppers who are hurting your profits today. This is why it is important to keep the long-term, big picture in mind, which you review on your Performance Platform. The sooner you let the technology run on your website and the sooner you begin engaging with your shoppers and offering customer service around sizing, the sooner you maximize results.
Need help? Contact us at email@example.com