Artificial Intelligence (AI) is making its way into the fashion industry, helping companies cut lead times, innovate, and manage their supply chains better.
Recently, researchers at the University of California at San Diego partnered with Adobe to create a new fashion-oriented AI. Not only will it learn about your personal fashion style, but it can also make suggestions based on your preferences.
There are many potential use cases for artificial intelligence; that much is evident. While there is plenty of focus on AI disrupting automation and even finance, Adobe has a very different plan. More specifically, it wants to help people upgrade their fashion sense by offering visual improvements to their clothing style. This will be done by teaching the AI about the end user and making suggestions based on his or her preferences. It's an interesting option well worth exploring.
For retailers, this particular AI will have many benefits. Creating personalised pieces of clothing for different types of consumers could make a lot of retailers more appealing to the masses. Moreover, such technology could help predict broader fashion trends in the coming years.
According to researchers at UC San Diego, there are two different algorithms at work powering this AI solution. A convolutional neural network learns and classifies user preferences for specific items. That is done by scraping Amazon for one's purchasing data in six different categories, including footwear and pants. This is not entirely new in the world of online retail, as models like this one have existed for quite some time now.
On top of this algorithm is a generative adversarial network, which is a type of AI that generates realistic images. It's a vital technology when it comes to, say, suggesting new types of clothing to consumers. This is an interesting dual-pronged approach to making people more fashion-aware while allowing them to do their own thing. The latter technology is not new, as generative adversarial networks (GANs) were invented a few years ago.
This new approach to deciding what to wear could have a major impact in the fashion world. Online retailers should be able to figure out what customers want beyond items that already exist. There is no reason to force consumers into buying items they may like in design, but would love to see in a different colour.
At Myntra, machines tell designers how to make clothes
Myntra's Rapid platform uses artificial intelligence to shrink the manufacturing process from 180 days to less than 45 for its fast fashion products.
In October 2016, Yash Kotak, Rohit Chauhan and other members of online fashion retailer Myntra's ambitious Rapid technology project hit a roadblock. The aim of the project was to deliver fast fashion products to the market as quickly and cheaply as possible.
Essentially, the plan would future-proof its business to understand demand, respond quickly to new fashion trends, cut costs, and reduce discounting in a fast and cost-effective manner. To realise these goals, Myntra started experimenting with artificial intelligence (AI) systems that recognised shapes, patterns and colours to produce garments that met popular demand at a speed that would be impossible for traditional apparel makers to match.
Initially, Myntra's plan was to identify underserved spaces, quickly make small batches of products according to the latest trends in those spaces and sell fast.
The Rapid tech platform analysed sales data from Myntra and Flipkart websites (Myntra merged with Flipkart in 2014), as well as the latest fashion trends collected by trawling Instagram, Pinterest, fashion magazines and similar sources. Rapid's technology would come up with a list of attributes: types of collars, sleeves, cuts, colours and other features currently popular with customers. Here was fashion via engineering.
The project was doing reasonably well. Based on the platform's analysis, Myntra's fashion designers had created a brand called Moda Rapido (Spanish for fast fashion). But this method was slow by the standards of the tech world. The machine would throw up suggestions in text form after crunching massive amounts of data, a process that took weeks. Moda Rapido, which was first introduced by Myntra in September 2015, was bringing in sales of Rs 2-3 crore every month, but wasn't scaling fast, not by the standards of tech.
Only a few months before this, Myntra's chief executive officer (CEO) Ananth Narayanan was pushing the Rapid team to experiment as wildly as they wanted. Rapid was one of the moonshot bets at Myntra and it didn't cost the company very much either. At the time, it only had 20 full-time staff, informally led by Kotak, a 25-year-old product expert and former entrepreneur, and 47-year-old Chauhan, another product expert and former entrepreneur.
Start to finish
In October last year, the Rapid team started working on an idea that had been explored before without success: what if you automate the fashion design process from start to end? If machines could recommend clothing attributes that customers currently favoured, shouldn't they be able to combine those attributes into popular final designs? The Rapid team started with T-shirts, because of their relative simplicity.
But the engineers knew the same approach would not work. Instead of asking their computers to suggest T-shirt attributes in text, the Rapid team wrote algorithms that would get their machines to produce images of T-shirts based on the discovered favourable attributes.
This did not work at first. The Rapid computers were unable to get a clear picture-quite literally-of the images they were crunching, so their output was abysmal. The images being fed to the machines needed to be of a far higher resolution. When the engineers increased the resolution of the images, they found their machines didn't have the computing power required to process such heavy images. Kotak and his team purchased graphical processing units (GPUs), ultra-powerful machines that are used in AI-related work.
The new machines started producing designs. The first ones were still hideous. The engineers tweaked the algorithms and, a few iterations later, there was improvement. They repeated the process: create the algorithm, keep feeding image data to the GPU, whet the new design, tweak the algorithm accordingly.
In January, they thought they had a set of some 30 presentable T-shirt designs, created wholly by the Rapid tech platform.
"Before October (2016), we were calling out everything by text because we wanted to give directions to the designers," said Kotak. "When you want to call things out you have to identify them. But when you use images you don't have to call things out. The machine can understand the images in an abstract way, understand the attributes by itself and incorporate them into the design."
The Rapid team convinced their bosses to take the machine-designed T-shirts to the market. In less than five weeks the products were ready. They went live at the end of February. These fully automated designs were created even as Moda Rapido designers continued to make products based on the recommendations by the Rapid technology. The question was: which would sell better, the machine-augmented designs or the machine-created ones?
"We debated whether we should tag them as machine-generated, but we decided not to," said Ambarish Kenghe, chief product officer at Myntra. "Only one product manager and I were aware which designs were machine-generated. We didn't want Myntra employees buying everything just because they were excited about the machine-generated ones."
It turned out that one category of the machine-created designs outsold the comparable category in each of the other 12 private brands owned by Myntra. Some of the other machine-created designs sold poorly, but the company knew they were onto something really big.
By this time Myntra had added categories such as trousers and kurtas to the Moda Rapido catalogue. It also started working on another fast-fashion brand, Here & Now. Here & Now offers similar products to Moda Rapido, only slightly cheaper. Most Here&Now products sell in the price range of Rs 400-1200. As the technology behind Rapid improved, both Moda Rapido and Here&Now began to offer more complex designs, such as printed kurtas and T-shirts, torn jeans and checked shirts. There's currently a limit to the variety of designs these brands can produce-the technology hasn't reached a point where it can produce things like T-shirts with complex graphics or shoes or accessories. In addition, when it is working all by itself the platform can only design T-shirts; all the other types of clothing are still put together by designers with the help of machines. But Rapid is improving fast.
In the pink
By September, Moda Rapido and Here & Now were together bringing in sales of Rs 12-13 crore every month, making them among the top 15 brands at Myntra. The two contribute 12.5% of Myntra's private brands sales, said Narayanan. Moda Rapido and Here&Now are Myntra's most profitable brands, generating net margins that are around 10 percentage points higher than the others, he added.
The Rapid team realised the vision of Gautam Kotamraju and Ganesh Subramanian, two former senior executives who started the Rapid project in early 2015 before leaving the company less than a year later. Kotamraju and Subramanian — the first is a fashion designer and the second a fashion veteran-wanted to create a fast-fashion brand that was data-driven. They sought to get products in customers' hands within 45 days of sending the final designs to the factory (this process typically takes more than 180 days). They also wanted to see if fashion designers could be wholly replaced by machines.
Only two years after they started the project, Myntra had successfully launched products that had been turned around in less than 35 days, masterminded by machines and engineers who knew very little about fashion.
"The core of the Rapid idea was to do fast fashion in an intelligent way, given that we have both a tech and a fashion DNA," said Kenghe. "In the initial phase of the project, there was less machine and more designer-input. Over a period of time, there's been more machine and less supervision."
Creeping to creative
Fashion is hardly the only creative field where technology is being used to transform the design process. Machine learning and older technologies are being used to varying degrees in fields such as architecture, car design, website design and even in making music. For instance, an AI-based platform, Shimon, developed at the Georgia Institute of Technology in the US, is already composing new music. Platforms such as The Grid and Firedrop are automating website construction. The stated objective of Google's Magenta project is: "Can we use machine learning to create compelling art and music? If so, how? If not, why not?"
In the world of fashion design, several companies and start-ups have tried to automate design, particularly in the last year, as AI technology improves by leaps and bounds. Some fashion brands are experimenting with IBM's supercomputer Watson and Google's technology tools to create designs. StitchFix, an American online retailer, has a tech platform called Hybrid Designs similar to Myntra's Rapid that automates fashion design using AI. Dozens of AI start-ups in fashion, including Stylumia, launched by Subramanian, the same Myntra executive who helped start Rapid, are attempting to do what Myntra is doing and more.
But Myntra has gone further than the others in applying technology in fashion design: it seems to have figured out a solid process. Broadly, there are three reasons why Rapid is ahead of other similar efforts. Myntra, which was started in 2007, has lots and lots of sales and browsing data. Apart from its own platform, the company has access to customer data on Flipkart and Jabong. Together these three platforms control roughly 70% of all online fashion sales in India, according to Myntra estimates. Two, Myntra has the cash to invest in AI technologies, such as GPUs, and attract top engineers. Three, it has built expertise in the supply chain side of fashion and has close relationships with suppliers and manufacturers because of its large private label business, launched in 2012. Myntra is able to convince suppliers to shrink the manufacturing process from 180 days to less than 45 for its fast- fashion products. Many companies could achieve one or two aspects but few can combine all the three.
"What we're doing is not just a technology solve or a supply chain solve," said Rajesh Narkar, vice-president of Myntra's private labels business. "It's about disrupting the way brands are built. Today, we've reduced the (manufacturing) process to less than 45 days. The idea is to do it even faster. We're asking ourselves: if there's customer intent to buy something, can we serve it then?"
"Lots of good data with mediocre (algorithmic) models will do better than great models with lesser data," said Chauhan. "That's the universal fact of AI. No one else has the amount of data we do. Let's say other start-ups build models that are better than ours. Do they have enough traffic to monetise that? The millions of users we have help a lot. And the third part is, how fast can they iterate on the supply chain. Manohar Kamath, head of Myntra's private label business, says that 'no designer job is going away' because of machines designing clothes. “But I will make them super-efficient. So it's a mixture of 90% system & 10% designer input,” he says. Yet, perhaps most crucial is the sustained commitment. Myntra has been working on the Rapid project for nearly three years. Its persistence is finally being rewarded. Given the nature of technology and AI, it's likely other companies will figure out automation in design, but Myntra has an important headstart over its rivals. "The last innovation in fashion was Zara with its supply chain model," said Narkar. "The next innovations will be led by AI. We're just off the track first."
Top to bottom
Apart from the financial impact, Rapid's technology is transforming the way Myntra works. The company has decided that all new private brands will be created using Rapid. Though Myntra can't do away with designers, the fact that all of its new brands will be machine-augmented to a large degree is a radical change in itself. The company is exploring launching brands in lingerie, watches and home furnishing based on Rapid.
Myntra is using the Rapid platform in other parts of its business apart too. Product descriptions on its site, thus far written entirely by content writers, are gradually being automated with the help of Rapid. It has created a Trends Store on its app that curates the latest trending products and shows them separately to fashion-conscious customers. In addition, the company buys products from third-party brands based partly on the recommendations of the Rapid platform.
"Now, we send an engineer and a buyer to go to brands (for sourcing products)," said CEO Narayanan. "Many of our brand partners send us a list of products in advance. The engineer crunches the data to figure out the probability of how much what will sell. We use that to decide how much and what kinds of product we should buy."
Myntra now plans to start selling the software platform as a service. It is in talks with more than five retailers and brands to sign up to the platform. It declined to name the retailers but said that the partnerships will be finalised by the end of January. The company has appointed Anurag Asthana, vice-president of product development, as the project's business head. Within the next six months, Myntra plans to expand the Rapid team to 60 people from 35, half of whom will be technology and product experts.
Next big thing
While Myntra has the edge over rivals in design automation, it's hardly an insurmountable lead. The company uses a combination of open-source tools such as Google's Tensor Flow and Caffe and then builds its own technology on top. Other start-ups can do the same. Myntra's attribute-extraction method to automate design works, but it's not something that scales up very fast. Its method has limitations in terms of the complexity of fashion it can design-remember, currently it can only create T-shirts by itself. The attribute-extraction method is also not easily replicable across categories. For instance, though Myntra successfully created T-shirt designs, it can't use the same process for jeans or dresses. From a technological standpoint, this a backward method: it favours data analytics over machine learning, which is currently the most widely used form of AI. So it's no surprise that the company's engineers are constantly experimenting with new technologies.
In October 2016, when the Rapid team started exploring the idea of generating completely automated designs, some of its engineers came across a technology called GANs. GANs were invented by Ian Goodfellow, a researcher at Google.
Neural networks, which comprise connected transistors that replicate the structure of neurons in the human brain, are a fixture in AI. GANs go further. They comprise two neural networks. In Myntra's case, for instance, one network comes up with the designs after being fed data. The other network critically vets the designs and gives feedback to the first network on where it's going wrong. Through this interplay, the two neural networks are trained to think like humans.
For nearly one year, Myntra's Rapid team trained and tweaked the GANs. They consulted with Goodfellow on how best to use this nascent technology. Finally, in September, the Rapid engineers presented new designs that have been developed using GANs. The new designs are expected to go live before the end of this year. If the designs work, Myntra could exceed the limitations of the attribute-extraction process and take its next leap in design automation.
"When we went deeper we realised that if we do it the attribute way, then it's going to be a never-ending process," said Chauhan. "There are finer and finer attributes, and it was becoming increasingly challenging to identify finer level details. It wasn't scalable. GANs is a more holistic approach. If we crack the problem once, we should be able to translate it to other categories fairly quickly. We're still using attributes, but as we develop GANs further, we will probably reduce the contribution of the attribute approach. Even so, we keep trying different things. GANs look promising today but a year from now, we may hit a roadblock. Some other technology may emerge."
Myntra faces a familiar adversary in its use of GANs: Amazon. In August, the MIT Technology Review magazine reported that Amazon US is working on automating fashion design using GANs. Amazon, which is expected to become the biggest retailer of fashion in the US online or offline this year, has all the things working for Myntra: data, tech and supply chain expertise. But Amazon is far bigger and far more accomplished at technology and supply chain management.
What about the designers?
The Rapid project has been a boon for the company in many ways. It has resulted in the creation of two brands at the cutting edge of tech in fashion. It's changed the way the company looks at creating new brands (though Myntra insists it will keep the contribution of private brands at 25-30% of overall sales). It has also given Myntra a potentially lucrative new business. From a sales, margins and valuation point of view, the Rapid project can deliver massive benefits over time. Inside Myntra, the Rapid project has created ramifications that are not immediately visible. One of the reasons the company is successful is that it has conveyed an image of being a fashion destination to customers. That was possible because the company hired fashion experts and gave them freedom and power. It combined the expertise of its fashion executives with that of its engineers. With the evolution of Rapid, it's clear that the role of fashion experts is reducing. At the very least, it's changing. Even the company's fashion experts, such as Narkar and Manohar Kamath, former chief operating officer at retailer Shoppers Stop, speak in tech terms. Dozens of content writing and other jobs have become redundant because of Rapid and other technologies, and dozens more will follow as Myntra improves its automation ability. The company may not fire designers, but it also won't increase the size of its design team. Then there's the question: how will the nature of the fashion design job itself change?
In other creative professions, such as architecture and music, research shows that as the use of technology increases, the nature of the creative jobs changes. While technology expands the boundaries of what can be done, in many ways it reduces the role of the creator.
So could the fashion designer role at Myntra become more of a quality assurance function in practice? Will the main responsibilities of a designer be restricted to ensuring that the right data is fed to machines and to vet designs generated by the machines?
"No, designers will start doing very different kinds of work as we move forward," said Kamath, head of Myntra's private label business. "What kind of inputs can they give the machine to improve? They have to continuously teach the machine how the human mind works."
What happens when the machines learn how designers think?
"What we're asking is, can we manage that with the same number of designers we have?" continued Kamath. "No designer job is going away. But I will make them super-efficient. So it's a mixture of 90% system and 10% designer input. While we are using machine intelligence, we are also using a finer layer of human intelligence to make it unique."
Alibaba Has Introduced An AI Fashion Assistant
In an age where offline sales are declining while online is experiencing growth, it looks like one form of shopping in replacing the other. Alibaba's AI fashion assistant thinks differently. It brings both worlds together by helping shoppers append in-store finds with what they can find in digital catalogs.
FashionAI, as the assistants are called, were installed in 13 shopping centres in China during Singles Day. To recall, Singles Day was a nationwide event created by Alibaba which successfully reigned in US$ 25.4 billion in sales online and offline.
Clothes in the stores were embedded with chips that allowed the digital screen to identify them. When a person walks into a fitting room fitted with a screen, the system immediately identifies the pieces of clothing and starts suggesting similar or complimentary pieces. Ideally, FashionAI will help the shopper find more options both online and offline. If a choice is made, a simple touch on the screen can summon an assistant to bring the desired item for a fit.
A neural network used in FashionAI helps it understand user preferences and make its suggestion-making skills better each time. Alibaba's Fashion AI is a strong display of how online and offline can work together instead of competing to create an even more immersive experience for the consumer. Consumers can physically fit garment while artificial intelligence provide the most sensible options to complete a look.
Smart displays in store have been around. Mirrors for make-up simulation at Sephora shop offer a glimpse into technology-driven brick and mortar stores. Where Alibaba stands out is bringing the capabilities of e-commerce directly to an offline shopper. This commitment will, in the future, not just come on Singles Day or any other special day. Alibaba has planned to build a physical mall which it will have absolute leeway to explore technologies that merge both offline and online shopping. According to a Business Insider report, the Alibaba mall will feature tech-based grocery stores and virtual fitting rooms.
Asics Leverages Artificial Intelligence to Avoid Holiday Returns
Returns are always a point of concern for retailers – and this rings especially true during the holidays. But some companies are taking actionable steps to ensure that returns are minimised. Case in point: Asics' partnership with True Fit. Asics uses True Fit's connected data platform, the Genome, which collects style attributes, returns data and garment specs. According to a spokesperson from True Fit, these features have enabled Asics to provide shoppers with personalised recommendations on what garments – and shoes – will best suit their performance needs.
Top Sneaker Brands Sound Off on Extra Butter's Legacy
What's more, the platform curates items based on these preferences, the spokesperson continued, which makes it easy for the shopper to find products they will appreciate – and keep.
As a result of the partnership, the spokesperson affirmed that Asics has increased conversions by 150%, reduced size sampling (from about 30-50%), as well as has encouraged shoppers to try out new items. Customers with True Fit profiles, the spokesperson explained, keep about 20 percent more of what they buy.
The numbers speak for themselves, but Jason LeBoeuf, director of e-commerce at Asics, confirmed the positive impact the partnership has had on Asics. "We are working to get more and more of our consumers signed up with True Fit so they can experience a personalised shopping experience online at Asics.com," he said. Romney Evans, co-founder of True Fit, said that personalisation is critical in today's market. "We're increasing the user's chance of success in finding something that works for them before moving on to something else. Fundamentally, that's the reason why personalisation is so important," he said. "If you can leverage data for personalization and increase confidence in consumers, such as through using True Fit's Genome Platform, consumers will respond by purchasing more – and returning less."
Stitch Fix Blends AI and Human Expertise
Stitch Fix provides a glimpse of how some businesses are already making use of AI-based machine learning to partner with employees for more-effective solutions. A five-year-old online clothing retailer, its success in this area reveals how AI and people can work together, with each side focused on its unique strengths. The company offers a subscription clothing and styling service that delivers apparel to its customers' doors. But users of the service don't actually shop for clothes; in fact, Stitch Fix doesn't even have an online store. Instead, customers fill out style surveys, provide measurements, offer up Pinterest boards, and send in personal notes. Machine learning algorithms digest all of this eclectic and unstructured information. An interface communicates the algorithms' results along with more-nuanced data, such as the personal notes, to the company's fashion stylists, who then select five items from a variety of brands to send to the customer. Customers keep what they like and return anything that doesn't suit them.
Stitch Fix's approach illustrates three lessons about how to combine human expertise with AI systems. First, it's important to keep humans in the business-process loop; machines can't do it alone. Second, companies can use machines to supercharge the productivity and effectiveness of workers in unprecedented ways. And third, various machine-learning techniques should be combined to effectively identify insights and foster innovation.
As research conducted across industry and academia shows, companies have an unprecedented opportunity to tap ongoing advances in AI and machine learning research to reinvent business processes. For instance, in analysing a five-year sample of almost 1,150 papers, we identified at least 12 techniques, visible in the chart below, that can be readily applied and combined with each other within a process. Stitch Fix is already applying several of these machine learning techniques in service delivery and R&D – and other companies can follow its lead.
Some extremely successful companies have made great use of recommendation engines to boost sales or improve customer satisfaction. When it comes to recommendations, is there room for improvement in the way Amazon and Netflix operate?
Stitch Fix, which lives and dies by the quality of its suggestions, has no choice but to do better–but it can't rely solely on machines to do this. The company collects as much information about a client as it can, in both structured and unstructured form. Structured data includes surveys with personal information such as body measurements and brand preferences. Unstructured data can be derived from social media accounts, such as Pinterest, or through online notes from people about why they are buying new clothes, such as a special occasion, a change of season, or because a certain new style caught their eye.
The automated recommendation system is at its best when dealing with structured data. But to make sense of unstructured data, people and their judgment are needed. Say a client wants a new pair of stylish jeans, an item that's notoriously tricky to fit right to a person's measurements. To start, the algorithm finds jeans (across a range of fabrics, styles, and even sizes) that other clients with the same inseam decided to keep – a good indicator of fit.
Next, it's time to pick the actual pair of jeans to be shipped. This is up to the stylist, who takes into account a client's notes or the occasion for which the client is shopping. In addition, the stylist can include a personal note with the shipment, fostering a relationship, which Stitch Fix hopes will encourage even more useful feedback. This human-in-the-loop recommendation system uses multiple information streams to help it improve. The algorithm absorbs feedback directly from the client – whether or not she or he (the company added men's options in late September) decided to keep an item of clothing. And the stylist improves and adjusts based on cues gleaned from client notes and with insights from previous interactions with the customer.