Can Amazon Use Machine Learning to Take On Fashion?

Amazon is taking on the fashion industry by developing its own collection of private label brands.

The Top U.S. Apparel Retailer

Think of the fashion industry and name the top five players that come to mind. Did Amazon make your list? Probably not. And yet, Amazon is on track to become the largest seller of apparel – in terms of both online and offline sales – by next year [1]. While traditional retailers such as Macy’s stand in shock watching their revenues decline, Amazon and other online retailers are swooping in to capture market share. Amazon is working to grow their apparel business beyond basic commodities like t-shirts and into trendier pieces like dresses. In order to approach this new segment successfully, they need buy-in from customers like you. How soon will it be until Amazon becomes the powerhouse behind your go-to clothing brands?

Amazon has the potential to double their apparel business. Source: Financial Times.

Developing Better Product

Amazon continues to enlist many top fashion brands as first party sellers – Nike, Levi’s, Stuart Weitzman. But, it has been a slow and uphill battle. Brands are hesitant to sell directly through Amazon, favoring tight control over their brand experience [2]. In response, Amazon has shifted its focus to developing its own product. They are investing to represent something larger in the fashion space: they want to be the company behind the next generation of brands you love, not just the platform. They have been developing and refining their own portfolio of 60+ private label brands [1, 3]. To do so, they are betting on their ability to harness machine learning to help create beautiful, sell-able products. Two things have made leveraging machine learning possible: 1) the massive amounts of fashion images and 2) the role of social media.

The most immediate result from their work is the ability to determine which images are “stylish” – a difficult endeavor especially given how subjective and hard to quantify “stylish” can be [4]. In a world where data is seemingly endless, Amazon is able to use artificial intelligence to cherry-pick appropriate images to track trends and use them as a source of inspiration for their own product development. What sets their technology apart is that they can apply it to a wide variety of data including unlabeled images – this means that they can utilize social media platforms such as Instagram and Pinterest [5]. This algorithmic approach gives Amazon an edge since they can learn a lot about their customers preferences via social media. Other companies such as Stitch Fix are similarly leveraging social media to better serve and predict customer preferences.

In a longer-term play, their research center, Lab126, is building an algorithm that will be able to design new products from trending images. Essentially, Amazon is creating an artificially intelligent designer in a move which could displace a traditional Creative Director or in-house design team. The algorithm uses generative adversarial network (GAN) to decipher a particular style by analyzing multiple images. It then uses inverse mapping to reconstruct new images that are in the same style. The simplest way to think about GAN is as a game whereby the generator G needs to create fake images to fool the discriminator D whose role is to determine which images are real and which are fake [6].

Machine Learning as a Reactionary Tool in Fashion

Machine learning can play a crucial role in product development at fashion companies. It can help identify trends which in turn increases the likelihood of a product’s demand and ultimately reduces product waste from overproduction. In a way, this new technology mimics how fast fashion disrupted traditional retail years ago, but takes it one step further. Instead of creating new trends, which is an expensive and unpredictable process, fast fashion adapts existing designs and sells clothing at a fraction of the price in record time (in short, what makes it all possible is shortening the production lead time from 12 months to just a couple of weeks). With machine learning, the process of identifying trends is shortened even further, giving companies like Amazon additional headway to create sell-able products.

Key Considerations

One downside, however, is that Amazon’s application of the machine learning process is reactionary. Instead of creating new trends, it helps identify which styles have taken off. It is important to note that AI has never successfully created something completely “new”. As Tim Oates, a Professor from the University of Maryland, posits, “People innovate in areas like music, fashion, and cinema. What we haven’t seen is a genuinely new music or fashion style that was generated by a computer and really resonated with people” [4]. To that end, can Amazon successfully create brand identities by using AI to identify ongoing trends? To what extent do brands need to create newness?

(772 words)

[1] Cathaleen Chen. Amazon Primed for Gain with Private Label Apparel. Business of Fashion (July 23, 2018).

[2] Lauren Thomas. Amazon Hopes to Ride Its Sales Momentum to Become the Top Apparel Retailer in the US. MSNBC (July 16, 2018).

[3] Eugene Kim. Amazon Owns 7 Private Label Fashion Brands. Business Insider (February 22, 2016).

[4] Will Knight. Amazon Has Developed an AI Fashion Designer. MIT Technology Review (November 12, 2018).

[5] Will Knight. 35 Innovators Under 35 – Inventor Ian Goodfellow. MIT Technology Review (2017).

[6] Alexander Lorbert, Nir Ben-Zvi, Arridhana Ciptadi, Eduard Oks, Ambrish Tyagi. Toward Better Reconstruction of Style Images with GANs.  KDDW (2017).

Previous:

Boeing: Additive Manufacturing Adding Enough?

Next:

Healthcare Social Networks: Can Crowdsourcing Patient-Reported Data Improve Outcomes?

4 thoughts on “Can Amazon Use Machine Learning to Take On Fashion?

  1. Very interesting article! This is a definitely expansion of our GAP case in marketing. I like that you took a more in depth look at the problem. I concur that Amazon will never be a ‘fashion leader’ however the opportunities for fast fashion are significant. I expect that Amazon will take advantage of its massive customer base and develop core products. This will be an interesting storyline to follow!

  2. Very interesting article. I find is fascinating how Amazon has taken over the US retail industry, and made retailers like Macy’s lose large amounts of revenues in a short amount of time. Retailers in Sweden are now fearing the entrance of Amazon later this year, as the move is expected to shake up the retail space in Sweden significantly. Still, despite the large number of consumers visiting amazon for their daily purchases, fashion brands have been resistant to enter the Amazon space due to risks of losing control of brand.

    While Amazon’s move to creating private label brands in fashion is an interesting response to the inertia of fashion brands, I have my doubts. In my opinion, fashion is very different from daily usage products. Just yesterday I was happy buy a yoga mattress from Amazon, but I would never even entertain the idea of buying a blazer from Amazon. Success in fashion is as tightly linked to the brand, especially the emotional connection to the brand, as the actual designs of the fashion pieces. I doubt that Amazon will be able to create a strong brand in fashion, especially as you sell cleaning products next to it.

    Another concern of mine, is the design process where an artificially intelligent designer replaces the traditional creative designer. As you mention it is reactionary, i.e. not creating trends but simply following them. Could Amazon risk being too late, and simply being seen as a copy-cat? However, I do believe that both science and art are necessary to create successful fashion pieces in the future. Social media has enabled companies to get an instant temperature check of what people believe is “in style” at the moment. By leveraging that coupled with fast lead times as well as a bit of art from the designer, I believe Amazon could create very successful designs. However, the question remains whether Amazon can create the necessary credibility as a brand to support the designs.

  3. I could not agree with you more that Amazon will never be anything but reactionary in using machine learning to create products. Using machine learning would only ever enable Amazon to become a leader in fashion in terms of quantity sold, but never in terms of shaping the next look or defining fashion trend. I do not think that Amazon could ever become a real threat in terms of quality or design to established fashion brands that have been known for their superior fit and fashion-forward thinking. However, Amazon can be very successful just becoming a mass fast-fashion retailer. In the end, I think Amazon’s play is more about reach and profit rather than gaining a name in the fashion industry.

    The part that I find most interesting is the point you raise about Amazon being able to use the algorithm to potentially decrease waste in fashion. Fast fashion has caused immense waste and contributes hugely to global warming. If Amazon could find a way to more accurately produce the right quantity of items, this would be a huge savings driver for them and it would benefit the world greatly. I wonder if there is a business model in which Amazon could sell this prediction service to other fashion retailers and try to leverage this advantage, while also doing something good for the planet.

  4. Great post! This is a great segway from the MKT case we had on Amazon. Your article certainly puts Amazon as one of the key players in the fashion industry, and I found it particularly interesting how you highlighted Amazon’s development of 40+ private labels. I’m wondering if algorithims or ML can be used to identify which trends before they become mainstream? Much research has been done on how trends from previous eras repeat themselves in modern day. Could Amazon identify such trends, and if so, could it develop it as a competitive advantage vs other non-tech fashion firms? Thank you for such a thoughtful post.

Leave a comment