How AI is making its way into Japan’s fashion industry
Recognizing what we know as “complex feelings” is an area that AI has not quite yet mastered. So having AI analyze the emotional and psychological processes that lead consumers to purchase fashion or beauty items is not something we can expect to be possible anytime soon. But two fashion service companies are nevertheless beginning to take on this challenge.
Rent-a-luxury-bag
Laxus is a service for renting bags with a monthly subscription fee of 6,800 yen. And we’re talking luxury brand bags, such as Louis Vuitton, Chanel and Hermès. They have a total range of roughly 23,000 bags up for rent, and by becoming monthly members, users can rent any bag without restrictions and use it for as long as they like, with no annual or delivery fees. The system has an exchange-format where when you want to borrow a new bag you’ll need to return back the one you currently have. And they also have a system where owners of luxury bags can entrust them to Laxus for leasing to others, and can receive remuneration when someone rents them out.
The service’s current continuation rate among members is 91.6%, and if you narrow that down to members who’ve used the service for over nine months it grows to 95%. They also have plans to expand overseas, and their recent venture in Manhattan, New York saw pre-launch registration surpass that of Japan three-fold.
Laxus is first and foremost a technology company. Their independently-developed AI is utilized in a number of different areas of their service, including in raising usage rate and continuation rate, and in improving their merchandise line.
Basically, Laxus trains its AI using a scoring system. If a bag is viewed but swiped along, that amounts to no points. Though if the bag is clicked on, that’s 10 points. And if it’s clicked on and also ends up being rented out by the user, that’s 20 points. By continuing to collect data like this, 24 hours a day 365 days a year, the AI eventually becomes able to predict the behavior of users, realizing what bags ought to be recommended to a particular user in order to create a higher likelihood that they’ll rent one. It can also group together a user’s favorite brands, and much more precisely recommend bags based off of user behavior.
However, AI still has its limits. It still cannot understand those vague areas of why users borrow particular bags. Was it because it was their favorite? Or was it a compromise? Users don’t want bags just because they’re selected for them by AI. In everyday life it’s normal to come across an item in a shop that for some reason you fall in love with and end up purchasing on the spot, however those type of subtleties of the human heart cannot be grasped by AI.
Fashioncoordinating AI
The other company we’re taking a look at is NEWROPE whichspecializes in AI that analyzes fashion photos to recognize what type of clothes the model is wearing and can search for and introduce similar-looking items. NEWROPE has teamed up with 300 models and Instagramers to develop and run the website and smartphone app, #CBK (Cabuki) that utilizes the AI in its featured fashion coordination photos. If there’s a style that you’re interested in, the AI finds similar-looking items of which you can purchase via partner’s online stores.
NEWROPE’s AI programs include Fashion Ojisan (or Mr. Fashion), which is the program that analyzes the fashion items within photos and introduces similar products and advice on how to wear them best. Then there’s also their AI program Mika which selects items to coordinate along with the items you already have. By indicating your taste, Mika can change its suggestions accordingly — for example, you might want something more girly, casual, or conservative — making it very similar to the attention-to-detail you get from a store employee.
For these types of AI, the goal is the conversion rate (CVR) — it’s all down to what needs to be suggested to a person who’s behaved like this in order to get them to make a purchase. It’s about finding what action will create the desired result. In this way, AI is well suited to the PDCA (Plan-Do-Check-Act) cycle of business.
What is definitive about AI is its ability to learn and become able to do things with more precision than humans can. However, being able to grasp preferences in design seems to be difficult for the time being. In other words, although one item is liked by someone but disliked by another, AI would perceive this situation as there being a gradation between “like” and “dislike”, and it would try and calculate how it could modify this attribute to increase the “like” as much as possible. At the end of the day, this is not really how the human decision process works.
Narrowing down enormous amounts of data
From the examples of these two companies, what can we see as the most optimal way of using AI in the fashion industry?
One way is to narrow down an enormous amount of items to a small amount, maybe 100 or just 20 items, and have users pick from the results. In the case of Laxus, 80% of members who withdraw from the membership state their reason as that the bags they want aren’t available, but the warehouses always have around 3,000 to 4,000 bags in stock, so surely there must be one bag they would like. You could say that users themselves are not properly narrowing down the items they’d like. This is where AI can help, and with further advancements in machine learning, AI’s precision in matching people’s preferences is likely to become more and more accurate.
Also, data suggests that showing items as part of a whole look, rather than just by themselves, tends to increase the conversion rate. Due to this, NEWROPE’s way of showing the whole coordination, and having AI handle the details for users, demonstrates how AI can support the way online stores market their fashion. And such a method looks likely to also have great potential in the beauty industry as well.
Translation: Ching Li Tor
Original Text (Japanese): Shidu Kumon
