Source: Amazon mobile app
Amazon taps AI to drive fashion recommendations
Amazon.com last week introduced StyleSnap, a feature that makes fashion recommendations based on user-submitted photos.
Users begin by clicking the camera icon in the corner of the Amazon app and then take a photo or upload a photograph or screenshot of a fashion look they like. The app returns recommendations for similar items on Amazon that match the look in the photo. Users are able to filter suggestions across brand, pricing and reviews.
“The simplicity of the customer experience belies the complexity of the technology behind it,” said Jeff Wilke, CEO of worldwide consumer for Amazon, in introducing the feature last week at Amazon’s re: MARS (machine learning, automation, robotics, and space) 2019 conference.
Indeed, a blog entry expounded on how StyleSnap’s deep learning algorithms are able to identify items within lifestyle images and indoor images with dim lighting and also classify those items into categories like “fit-and-flair dresses” or “flannel shirts.”
The technology can be “trained” to identify the outfits by feeding it a series of images.
Amazon wrote, “To have neural networks identify a greater number of classes, we can stack a greater number of layers on top of each other. The first few layers typically learn concepts such as edges and colors, while the middle layers identify patterns such as ‘floral’ or ‘denim’. After having passed through all of the layers, the algorithm can accurately identify concepts like fit and outfit style in an image.”
StyleSnap also solves the “vanishing gradient problem” when neural networks “will stall and eventually degrade after a certain number of layers have been added.” Residual networks are applied to allow the training signal to skip over some layers in the network. Amazon wrote, “A unique method developed by Amazon researchers allows the network to learn new concepts while also remembering things it has learned in the past.”
StyleSnap also promises to help fashion influencers expand their communities and supports Amazon’s Influencer Program through which Instagrammers and bloggers earn commissions by recommending Amazon items. Amazon also just teamed up with popular Instagram influencers to launch The Drop, a streetwear label.
Asos, Stitch Fix and Uniqlo are some other retailers promoting AI-driven style suggestions for customers.
Discussion Questions
DISCUSSION QUESTIONS: What do you think of StyleSnap and the potential of artificial intelligence to support fashion discovery? Is StyleSnap more about appealing to Instagram influencers?
Poll
BrainTrust
Anne Howe
Principal, Anne Howe Associates
Ken Lonyai
Consultant, Strategist, Tech Innovator, UX Evangelist
If Amazon can get this right, and that’s a big IF given the fickleness of the fashion world, it will be huge. Using artificial intelligence can help people better shop for clothes and “find that look” on the spot. Assuming the tech — and Amazon inventory — can back this up, it will certainly find followers.
The tech is actually quite fascinating, but knowing “the algorithm” is the fashion advisor is kind of off-putting in the context of fashion. Because fashion is so fickle and personal, maybe branding it like Amazon did with “Alexa” is something to consider.
Color me skeptical. I’m going to say that apparel shopping, and ultimately buying, is both a thinking and feeling experience. I’ll accept that algorithms can filter a broad array of choices and offer up an edited menu. So they’ve helped with the thinking. Final decisions are about feeling, both physical and emotional feeling. What is the weight, texture, wash and handfeel of the garment? How does is feel when tried on … fit? Sounds like algorithms can help save steps, but not necessarily help with final decisions.
This is a pivot from Amazon’s deal with Snapchat where they are basically doing the same visual search. Visual search has been around for some time and Pinterest Lens is a great example — the style snap is an almost identical match. The difference is adding the transaction process to the end of it driven to Amazon vs other sites.
The disadvantage for style snaps is that it’s based on pictures that you upload. For folks who like fashion, the collaboration and social value of searching through other’s pictures is missing — not just your own wardrobe or snaps you see on the street, it really widens the market. Not sure how it fits with instagram at this time, but there’s a segment that is regularly taking pictures with their smartphones and Amazon is great at expanding their market.
Machine learning from image training sets is not new. Whether Amazon has made real improvements that will allow it to train more accurately is an unknown, awaiting user feedback.
All of digital commerce is ultimately going this way. For example, someone that is digging a hole and snaps the shovel handle will be able to use a camera to capture the product and also say “How soon can I get a shovel like this for under $25.00 delivered to me?” Anyone that steps back and looks at all the pieces in play today can see that coming sooner than later.
Thanks for taking my head beyond apparel with a very clear example.
AI, deep neural networks, or any other touted new or old technologies all have a blind side. What they should be solving for is individual human sensory preferences. The #1 reason fashion items are returned? Individual sensory preferences issues of fit, look and feel. Solving for visually alike is old thinking and non-effective, parsing the complexities of human cognition into a non-human like solution. Just as useless as using collaborative filtering, segmentation inferred behavior, or wait for it — the weather — to recommend items people actually wear on their bodies.
Hence, even with all the “new” technologies since the beginning of online shopping for clothes [cira 1995] online apparel retail conversion rates are still 3% and with staggering return rates of 30% to 40%! Lost in the quest for new technologies, human cognition and human sensory preferences.
StyleSnap is just another technique based solely on the human-less limitations of computer science. What to solve for? Explicit [implied] versus Implicit [absolute] human behavior. The absolute dichotomy of computer science and human-centric abstraction. No wonder billion dollar verticals still have not crossed the sensory chasm between the physical and digital worlds. Solving for individual human sensory preference is the key.
Well done, very informative comments!
This has the potential to make shopping for clothes on Amazon easier, since a search for “women’s sweaters” may or may not turn up anything a shopper is actually interested in. Since its launch, Amazon has used algorithms to tailor recommendations to their shoppers, whether or not StyleSnap will do a better job of customizing clothing recommendations remains to be seen.
Amazon clearly isn’t the first to be experimenting with this sort of thing — it’s interesting in the context of whether this will help Amazon take sales from elsewhere. After all, it would be foolish to ignore the fact that Amazon has been reported as overtaking Google as number one when it comes to customers searching for products to buy. This tech could make that search process faster and easier — especially when it comes to seeing something and being inspired by it.
Obviously it remains to be seen how good the recommendations are, but if customers are seeing products elsewhere, searching for, finding and buying them through Amazon then that’s something to watch. I agree with the other comments though that visual search only solves one part of the online apparel problem with fit and feel being hugely important factors when it comes to keeping an item.
More and more we are seeing AI help with recommendations. Someone who shops online for fashion or uses other social channels like Instagram, Pinterest, etc. will enjoy this experience. If nothing else, it will help educate the customer about the latest and greatest. And, the technology will only get better.