Main areas of interest
We see five main areas of interest when it comes to machine learning in fashion.
Smart tagging of large catalogues and automation of the process. Giving products labels which are more complex than standard “long dress”, “long sleeve”, etc. would shorten the process of finding the product. Instead of standard filters/search, a customer could write in natural language whatever he/she wants (“dress for going out with friends on a Friday night”) and then specify his/her search with more words. The result: quickly finding products one is looking for. Implementation would involve visual recognition algorithms with a well-prepared basis of tags – all that being done via supervised learning at first.
Visual search. A customer provides a photo (say, of a celebrity wearing a particular dress) and demands to find something similar in a catalogue. Visual recognition algorithms are well-developed and their basic form can be easily implemented using ready-to-use tools (IBM Watson, Google Cloud Machine Learning, etc.).
Predicting trends, fashionability for next seasons. Every fashion brand wonders what will be trendy in the next season. Can it be foretold? To some extent definitely yes – it involves analysis of buying habits, seasons (Spring/Winter), geographical location, age, income and many other factors which influence whether a given product will be trendy. Big data at play! Moreover one can try to factor in the signals from social media and newspapers by looking through thousands of messages and photos of influencers.
Recommendation algorithms. Personalize approach to each user by analysing his/her choices, visited pages, social media. The more you know about your customers, the better service you can provide. But more data means also more challenges.
Online personal stylists. Chatbots can help clients make their choice, especially when they are integrated with smart tagging so that the client quickly gets to a product he/she wants. They are natural when one tries to import a personal shopping experience to an online world.
How it is done currently
Let’s have a look at already implemented solutions.
Stitch Fix analyses personal data of customers and then, together with a personal stylist, sends 5 products to a client. It’s a nice blend of human-AI interaction. The whole story is described in Harvard Business Review.
Edited is an example of data analytics company specializing in fashion. It analyses trends worldwide, monitor prices, shows history of previous launches of fashion brands. By tracking trends Edited can help launch a product in the right moment.
Net-a-Porter collaborated with IBM to create a visual recognition system where a customer can submit a picture of a celebrity and find similar products in a catalogue. IBM’s cognitive system was used (Watson Image Recognition). For a ready-to-use solution in fashion have a look at ViSenze. Upload a photo and it will show you products closest to those presented. Moreover ViSenze offers also automatic product tagging.
Big players like Zalando are also looking at machine learning. The general idea is that having huge amount of data regarding personal choices, one will be able to extract categories of clients and target them accordingly, so the overall buying satisfaction goes up (together with sales).
With the global fashion industry being evaluated at 3 trillion dollars, we are sure that there are more AI solutions to come. In fact fashion industry has just started adopting it. In order to develop further fashion brands have to start building/hiring data science teams to help them understand big data they have and turn it into profits.