ZOZO’S PLUNGE INTO MACHINE LEARNING
ZOZO, Japan’s dominant online retailer of apparel and related lifestyle products, with approximately $1B in sales and $7B in market capitalization, has taken on a journey to recreate the world of apparel to the next level.  It aims to do so in two steps: 1) enable mass customization of private-labeled apparel products, and 2) quantify all aspects of what makes styles “cool” to become the global leader in product design.
Machine learning serves a critical function for ZOZO, because it needs to predict and manufacture ergonomic, stylish products to satisfy each customer, based on given basic inputs around the customer’s body measurements. Originally an online retailer of apparel brands (think fashion-dedicated Amazon), ZOZO has risen to its heights by developing unparalleled strengths in user experience and product selection.  To take on the next challenge, the company has now introduced a just-in-time production for private-label brand products (named “ZOZO”) such as suits and jeans, such that it can create customized products within two weeks of order placement. 
ZOZO Online Retail Page
Launch of Customized, Private Label Brand “ZOZO”
Specifically, ZOZO relies on machine learning to substitute what had heretofore been done by experienced human beings: translating customer order specifications into physical patterns and defining corresponding manufacturing processes that balance design precision with order size, raw material yield, and speed.  The CAD program needs to define the optimum balance and translate it to the physical set-up of sewing machines. Despite the challenge, ZOZO stands in a unique position to leverage its vast data resource; in addition to its 23 million user information and outfit data, it is rapidly collecting data from production trials on body measurement data from over 500,000 ZOZOSUITs distributed globally (see next paragraph). 
IN SEARCH FOR QUANTIFYING STYLE
In the short term, ZOZO is very clear on its goal to becoming the world’s leader in mass customization of its private-labeled brand “ZOZO.” In the past year, ZOZO Technologies has made two attempts to bring this to reality, both with mixed results. First, ZOZO introduced a full-body suit, ZOZOSUIT, containing sensors of 150 varieties that collectively measure 15,000 different parts of human body.  When ZOZO failed to mass-produce the suit, it moved instead to a low-tech, dotted suits without sensors that could measure a human body using 360-degree selfie. In Oct 31, 2018, ZOZO abandoned the use of this suit entirely, claiming that it has “collected enough data to predict the needs of human body with simple information such as height, weight, age, and gender.” 
First ZOZOSUIT Launched
Second ZOZOSUIT Launched
In the long term, ZOZO is taking on an even more ambitious goal: “Quantify fashion, such that we can scientifically prove the ambiguous, captivating, and complex factors that makes fashion cool, cute, and beautiful.”  ZOZO has launched a separate company ZOZO Research to tackle this problem, actively recruiting experts in fields ranging from machine learning and pattern recognition to ergonomics and behavioral economics. In parallel, ZOZO is actively pursuing acquisition of external startups and ideas to accelerate the process.
OPPORTUNITY TO LEVERAGE THE “TRADITIONAL”
The mission of ZOZO is to “make the world a cooler place and thereby bring smiles to the world.”  ZOZO should consider making two critical adjustments in order to achieve its goal, both in the short term and the long run.
First, as ironic as it may sound, ZOZO needs to learn from its recent mistakes and make investments in hiring experienced designers and merchandisers who can fill the gap between the traditional manual production and ZOZO’s automated production. While I am optimistic about ZOZO’s long-term prospects, ZOZO should seek to accelerate its learning by working side-by-side with designers and merchandisers to articulate which factors make certain designs desirable and how to bring those designs to reality in (semi-)automated production sites.
Second, ZOZO should also seek to connect with customers in brick-and-mortar channel to receive feedback in the tangible space. Currently, ZOZO’s customer base is heavily centered around those in 20’s and 30’s who are completely tech-savvy. However, the target segment that really benefits from customization could be those in older age segments, who have higher economic freedom and are actively looking for apparel products that can make them look fit and stylish as they age. Building an equivalent of Nordstrom Local would enable ZOZO to gather in-person data that could further advance its pursuit of providing stylish products for the mass.
Nordstrom Local Store Display
CAN ZOZO WIN?
While ZOZO can leverage its presence in Japan to fuel growth into uncharted territories, startups such as Boston’s Stitch Fix seem to be ahead in applying machine learning to at least predict consumer demand. Simultaneously, Amazon also seems to be on the road to replicate its merchants’ products. Will ZOZO be able to drive the industry in offering customized yet stylish product for the mass? What capabilities will differentiate who will win in the long haul?
 “Company Information: ZOZO, Inc.” SPEEDA, 13 Nov. 2018, www.ub-speeda.com/company/
 “Reviewing the Path of King of EC Apparel: ZOZOTOWN.” WWD JAPAN, INFAS PUBLICATIONS, 13 Nov. 2018, www.wwdjapan.com/735174.
 “Learning from the Failures of ZOZOSUITS.” Commerce Online, shogyokai.jp/articles/-/1241.
 Fukachi, Masaya. “E-Commerce Expert’s Advice on How to Sell on Amazon, ZOZOTOWN, and Owned Websites.” Note, note.mu/fukaji/n/n662f4cfb26e8.