Data science challenges in a fast fashion apparel company
Written by Meigarom Lopes — Data Scientist at AMARO
Introduction
Data scientists can operate in a variety of different fields, but one condition is a must: data availability. They can work with finance, logistics, e-commerce, medical industry, insurance, among others — some of these segments have already been hiring data scientists to work on challenging projects for a while.
E-commerce, for instance, has a great data diversity. Customers express their preferences, tastes, styles, moods and desires to buy through data generated during website navigation, providing important infos such as history of items purchased, demographics, color preferences, platform/device used, products visited, clicked, liked, added, and so on. The fast fashion apparel e-commerce segment, in particular, is very rich in terms of data diversity and availability, mostly originated from the regular e-commerce operations, such as purchase activities and general fashion content, like the pictures of models wearing the products.
The main objective of e-commerce retailers is to create a memorable purchase experience for their customers. To do so, they need to recommend interesting products, help customers choose the next fashionable look, provide an intuitive search platform to browse products and eventually return them, predict the next product to be purchased, send relevant items to customers that might be interested, predict the growth of the company in the next few years and so on. On the other hand, designers and members of fashion product teams are interested in creating fast fashion apparel collections that will provide excellent fitting for customers, making them look wonderful and comfortable at the same time. In order to accomplish that, they rely on external data to predict a sort of things, from the next fashion trends to prices and the number of styles to be manufactured, among many others. From the data science point of view, this type of industry is a true playground that offers a number of challenging projects originated from both the e-commerce daily basis operations and the fashion content field. In this post, we will discuss the three biggest challenges of this amazing scenario.
How many items will be sold of this new product?
The first big question that retailers have is: how many units of this new product will be sold? It looks like a standard question in the e-commerce field and an easy job for those who domain a set of statistical and mathematical tools. At the first glance, as a data scientist, you could model this problem using Time Series techniques to predict the sales curve of the specific new product or you could use the Linear Regression Analysis to predict the number of sales in the next few days, for example. It may look like an easy job, however, the nature of the fast fashion business generates conditions that can turn standard problems into challenging ones. However, before answering the question above, data scientists have to take into account the following points:
First, the size of sales historical data. The life cycle of a fast fashion apparel collection is very short, as the name of the industry suggests. It takes about three to four weeks to design, manufacture and deliver a new product to the sales stock, moreover, the product won’t remain available for a long time. From a statistical modeling perspective, having only a few days of sales historical data to build a time series model may not be enough to capture the trend and the seasonality of the product sales curve, which are essential informations to make any prediction. Therefore, a short period of sales historical data is a something that data scientists usually need to address during statistical modeling, otherwise, the model might result in inaccurate result.
The second condition is the volume of products: it’s surreal in a fast fashion apparel collection, with a great variety of styles, with each one being composed of a combination of features like colors, sizes, designs, details and so on. It becomes unreasonable to build a Time Series model for every single product in a new collection. One possible solution is to group products and create a Time Series model for the entire category. However, the problem with this approach is the level of prediction. At a category level, sales predictions are not very useful, because they do not help to define the number of units of each product that needs to be produced. The best sales prediction would be at the product level, product by product. In other words, the challenge here is to predict the number of product sales without building thousands of models.
Lastly, designers want to predict the amount of product sales during the design phase. They want to know if the next best-seller product will be the one with more or fewer details, if the next best-clicked item will be printed, colorful or if the next most photographed, posted, reposted and commented product on social media will be either an audacious model or a regular one. So, the best scenario would be that fashion designers, during creation phase, try out all possible apparel features combinations to create a new product that will go viral among people.
How to determine how long will a product take to sell out?
The second big question made by retailers is: how long will the new high fashion product take to sell out? This time, the specific details are: the fashion world involves all kinds of people — from the ones that see the act of choosing an outfit just as a necessity to leave home, to the ones (such as celebrities, models and influencers) that have their outfit choice as part of their livelihood. Those who promote a certain fashion style play an extremely important role in fast fashion apparel e-commerce. They are usually followed by thousands or millions of people on social media and everything that they do, speak, wear, eat and play reach a huge audience that can be used as an important channel, working as a catalyst for sales. Depending on the reach of the influencer and how numerous their followers are, the sales curve can have spikes in a short period of time, leaving the stock for that one specific product empty overnight. Without any doubt, this external influence is a feature that clearly has a great impact on the statistical model and needs to be considered. However, the challenge here is regarding how to take this effect into account once it is hard to measure this potential.
How to measure beauty? What is visually appealing to people?
The third question involves discussing what is beauty and how to measure it. As we usually say, beauty is an individual perception that depends uniquely on the person and his or her references. For instance, while the same fashion model picture can be so beautiful to a person that make he/she wants to buy the product, others may not feel attracted to it at all. Of course there are tons of other variables that are taken into consideration by customers before they complete the purchase including price, delivery time, payment options, sizes and colors available, fashion stylists recommendations and monthly budget, but the combination of all these variables help customers decide. Clearly, modeling the will to purchase using all the variables available, including the more tangible ones and those that are abstract, is a really hard task. Even creating a ranking of fashion model pictures based on beauty can be very challenging. Feelings, impressions, mood, and other non-palpable factors make it hard or even impossible to communicate through data. For sure, this is one of the hardest challenges in the fast fashion industry.
Conclusion
As long as we have data available in any field, data scientists can work on great challenges, sometimes solving problems, others developing great technological feature. Each field has its own peculiarities, which can turn a standard problem into a really tricky one. Data scientists must be prepared to deal with these particularities and be able to suggest out of the box solutions in any context. Considering all the details involved, fast fashion companies are for sure the best places to find problems that will make you grow up in your career.
