The Profile Builder is an AI-based solution that allows customers to provide specific information on their style preferences. This information enables to create a personalized experience for the customers and provide more relevant recommendations. Since it is cumbersome for customers to browse through 1000s of items, so a small representative sample is selected to be used in the Profile Builder. The idea is which products should be shown to learn the most about the customer\u2019s preferences and maximize the value capture by increasing the chances for customers to purchase. Each product is represented in a vector obtained from an algorithm called collaborative filtering.<\/p>\n\n\n\n
Collaborative filtering learns a vector based on customers\u2019 interactions with products. Customers that purchase the same items often tend to have similar taste. Collaborative filtering takes into account the views, products added to bag or any other information that has high predictive power such that a machine learning technique known as clustering is implemented. Clustering performs a grouping of a collection of products so that they are in the same group (called a cluster) \u2014 items in a cluster are similar to each other, thereby transforming it into embeddings for each product and customer. A key property of these embeddings is that similarity in the embedding space (according to some similarity function) encodes similarity between products or a preference between users and products.<\/p>\n\n\n\n By creating a grouping of 36 different clusters, and one product per cluster is selected and a desired list of 36 products is presented to the customer based on their preferences. Through this method ASOS leverages machine learning to help its customers find their dream product by providing high-quality personalization through multiple touchpoints during the customers\u2019 journey.<\/p>\n\n\n\n Buy the Look (BLT) feature<\/strong> \u2013<\/p>\n\n\n\n ASOS took their personalization approach to the next level by developing an outfit recommendation tool. Instead of simply recommending similar items, the outfit generation tool recommends complementary items – a much more complicated task.<\/p>\n\n\n\n Combining items of clothing into an outfit is a major task in fashion retail, it requires an understanding of style, color, shape, pattern, fabric, fashion trends, customers\u2019 style preferences and the occasion on which the outfit will be worn. For example, if someone is viewing for party outfit, in the recommended products section the primary image customers might see matching high heels, evening bags and jewelry.<\/p>\n\n\n\n To tackle this ASOS incorporated a Machine Learning tool that consists of a multilayer neural network fed by visual and textual features such as images, text descriptions and categorical attribute to learn embeddings of items and compatible items of different types are embedded close to one another. In other words, it analyzes and groups together items based on attributes, which match stylistically and can be worn together.<\/p>\n\n\n\n The ML model is trained such that each training example consists of one outfit made up of a set of ASOS products. ASOS utilizes an internal Buy the Look (BTL) dataset which provides almost 600,000 outfits curated by ASOS stylists. The dataset is taken from ASOS product description pages so every product in the catalogue appears once as a seed product. New outfits can be generated by sequentially adding items and re-scoring i.e., each outfit made up of a seed product and a variable number of styling products (e.g., a dress may be styled with a pair of shoes and a bag). An outfit template is designed which is a set of product types with which to complete the outfit. The aim is to find the set of items of appropriate product types which maximizes the outfit score.<\/p>\n\n\n\n Challenge and the change ASOS had to make to address the challenge<\/strong><\/p>\n\n\n\n In the fiercely trend-driven fashion world, the company needs to produce data models as quickly as possible. ASOS adds about 5,000 new products every week to a catalog of 85,000 items as trend cycles are getting shorter.<\/p>\n\n\n\n ASOS ML modelling involved three specialized teams (Data science team, big data team & Inferencing team) working in three disparate data stacks. Waiting for one team to complete its part of a project blocked the other two, and end-to-end testing across the three stacks complicated the iteration cycles as models evolved. It typically took six months to bring a new model online. The main challenge is to accelerate this pace.<\/p>\n\n\n\n The ASOS teams wanted tools that would help them accelerate the modelling process and a new, enterprise-wide ethos\u2014breaking down barriers for teams and developing best practices for dealing with data science and modelling. By standardizing on Microsoft Azure Machine Learning service & by unifying the tech stack, thereby bringing ASOS engineers together, ASOS was able to reduce time-to-market for a recommendations model that increased marketing agility from six months to about six weeks. Additionally, it also helped data scientists to not to worry about ML infrastructure, and spend time focusing on their models and running experiments. <\/p>\n\n\n\n ASOS is a great example of a company that uses AI as an improvement of its business process and its relentless focus on customer. As these AI methods are set to be revolutionary in driving sales but also in decreasing the volume of returns for online retailers, which directly impact sales revenue and profit.<\/p>\n\n\n\n Source:<\/p>\n\n\n\n Reference-1<\/a><\/p>\n\n\n\n Reference-2<\/a><\/p>\n\n\n\n![\"\"](\"https:\/\/d3.harvard.edu\/platform-digit\/wp-content\/uploads\/sites\/2\/2022\/11\/EMBEDDINGS_1.jpg\")
<\/a><\/figure>\n\n\n\n<\/a><\/figure>\n\n\n\n<\/a><\/figure>\n\n\n\n<\/a><\/figure>\n\n\n\n<\/a><\/figure>\n\n\n\n<\/a><\/figure>\n\n\n\n