Fashion today is a $1.2 trillion industry globally. For an industry that has been around for the last 200+ years, it’s remarkably unchanged. Most of the fashion supply chain remains the same. From a design, usually inspired by haute couture on a ramp in Paris/Milan, to a design that is then interpreted for street fashion, the entire process of design to development to production to distribution is a 18 to 24 month cycle. Given the time and the inability of being able to predict trends so far in advance, almost 30 percent of the inventory gets liquidated in an end-of-season sale, which is a lot of waste in the system.
How can the industry solve this large and age-old problem? Given the recent data proliferation, the answer is staring at us: Could we use the (annonymised) data signals from the millions of customers who browsed various platforms to better predict what customers would want?
Several companies are now experimenting with algorithms and technology to help utilise this data meaningfully. Given advances in machine vision, a deep fashion taxonomy and plenty of data a machine can “learn” to very accurately predict trends (eg: colour, collar type, sleeve length) by each article type (eg: dresses, watches, shoes, T-shirts). This helps with more effective curation and selection for customers but still does not cut down on the waste in the system. The next step from here would be entirely machine-generated designs, and how that could positively impact both customers and the business.
Various machine-learning frameworks now exist that can consume this data (attribute-recipes) and spit out a whole set of machine-generated designs. Once we know the high probability best-sellers, one could take advantage of the manufacturing hub in India to create a really short supply chain and have these products on the shop floor in approximately four weeks. This could be further cut down if we applied automotive manufacturing techniques of configure to order (i.e. keep the base building blocks the same--fabric types, buttons, zippers, etc), which further helped reduce time to market.
The results are incredible. The machine-generated designs are better: Companies have seen 2x to 3x the rate of sale (including substantially higher sell thoughts for full-price inventory). Since these were small-batch productions, they had almost no waste whatsoever. If you combined this with the fast supply chain, and hence, the lower working capital (i.e. ~50 percent of inventory cover of typical brands), brands can potentially achieve 1.5x to 2x bottom line relative to traditional brands. The best part is that the machine continues to learn and get better with every new article type as it was a complete feedback loop.
I believe this is a very small beginning to how we can use data, algorithms and machine-vision technology to start to disrupt whats been traditionally a very 'art'-oriented industry. I do not for one minute believe that this will replace the genuine creativity or designers world wide--however, I do believe that having better data and tools will help them and help the entire industry reduce waste and deliver much better value and experiences to the customer.
This example is just in the design and some parts of planning. I believe that with advances in technology (IoT, 3D manufacturing and so on), every aspect of the fashion supply chain will be dramatically transformed in the next three to five years. We are already seeing driverless cars; can designer-less fashion be too far behind?
The author is a Former CEO of Myntra & Jabong.