Read ‘THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)’ in the attached file Complete the Case A

9- 5 2 1 – 0 7 0 R E V : M A R C H 1 , 2 0 2 1

Senior Lecturer Jill Avery, Professor Ayelet Israeli, and Emma von Maur (MBA HBS 2020) prepared this case. It was reviewed and approved before publication by a company designate. Funding for the development of this case was provided by Harvard Business School and not by the company. Emma von Maur is a former employee of THE YES. Certain details have been disguised. HBS cases are developed solely as the basis for class discussion. Cases are not intended to serve as endorsements, sources of primary data, or illustrations of effective or ineffective management.

Copyright © 2021 President and Fellows of Harvard College. To order copies or request permission to reproduce materials, call 1-800-545-7685, write Harvard Business School Publishing, Boston, MA 02163, or go to www.hbsp.harvard.edu. This publication may not be digitized, photocopied, or otherwise reproduced, posted, or transmitted, without the permission of Harvard Business School.

J I L L A V E R Y

A Y E L E T I S R A E L I

E M M A V O N M A U R

THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)

“It’s the holy grail of online retail: suggesting the exact item someone wants to buy before they even know it. Many start-ups have tried and largely fallen short.”

— A Business of Fashion journalist1

Co-founder and CEO Julie Bornstein (HBS ‘97) rolled up the sleeves of her KULE sweater, her most recent purchase from THE YES, a multi-brand shopping app she had launched with her co-founder and chief technology officer, Amit Aggarwal, in May of 2020. THE YES offered a new type of buying experience for women’s fashion, driven by a sophisticated algorithm that used data science and machine learning to create and deliver a personalized store for every shopper, based on her style preferences, size, and budget. When a woman downloaded THE YES app, she embarked upon an interactive shopping journey that leveraged a fun, easy, gamified user experience (UX), to collect a stream of data from her that could be used to dynamically curate an ever-changing product assortment personalized just for her. Pundits dubbed it algorithmic retail and intelligent e-commerce, but to Bornstein, it was the realization of her longtime dream to reimagine and enliven online shopping.

It had been a whirlwind four months since the app’s launch. Now that they were engaging with real users, they had data from their initial customers to decipher so that they could assess and optimize product-market fit. Before investing significant levels of paid media to acquire customers, Bornstein was searching for proof that the UX and personalization algorithm were working sufficiently well enough to deliver on the customer value proposition. Aggarwal was advocating for paid media investment sooner to bring more users in so the algorithm could learn and improve its performance.

The company’s investors were eager to see the team’s plans for expanding the user experience and further monetizing the platform. Several ideas were on the table, including the development of social shopping features to make the shopping experience more viral, the design of an influencer program to bring fashion influencer voices onto the platform, and the construction of a customer loyalty program. As Bornstein and Aggarwal debated how to allocate their resources, they realized that there were a lot of good ideas, but that they could not do it all. The 2020 COVID-19 pandemic had already delayed their launch by two months, so they were eager to get moving.

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

521-070 THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)

2

Company History In the fall of 2017, Bornstein, chief operating officer of Stitch Fix, an online personal styling service,

found herself at a crossroads. With Stitch Fix’s IPO prep in the rearview mirror, she questioned what she wanted to do next. She had spent decades driving digital transformation at various retail organizations. She had launched department store Nordstrom’s original e-commerce site and then served as chief marketing officer and chief digital officer at Sephora, a beauty retailer known for best- in-class e-commerce. The way she saw it, she had two career options now: revolutionize another incumbent retailer as CEO or strike out on her own.

Now felt like the right time to take the helm, as her deep understanding of online shopping had revealed significant shortcomings. Few e-tailers had ventured to change the aging e-commerce technology infrastructure that had been developed two decades earlier, which had been designed to transfer printed merchandise catalogs onto the web. Most fashion websites today offered basic categorization of items by category (e.g., jackets, dresses, pants), and allowed visitors to filter, sort, and search for items. For years, visitors had seen largely static landing pages and standardized search results. As technologies advanced, some retailers began to vary their landing pages based on the geography of the user, the browser and operating system being used, and clickstream analysis from shoppers’ previous visits. However, the degree of personalization based on expressed or implied preferences was still severely limited. In this aging infrastructure, the onus was on the user to scroll through page after page of items in an “endless aisle” to find what she was looking for. Bornstein noted, “For a lot of people, buying fashion online is a truly overwhelming shopping experience and we have the low conversion rates to prove it. It’s amazing how much data e-commerce companies are gathering on users and how little they are using it.”

Amazon best illustrated the issues about which Bornstein was so frustrated. By 2020, Amazon had become the largest apparel seller in the U.S., with over $30 billion in sales, and was poised to grow even larger.2 The e-commerce giant sought to do to fashion what it had previously done to all other industries: commoditize it. One of its key strengths, offering the broadest assortment that captured the long tail of consumer preferences, yielded a dizzying array of products, of varying quality and prices, without any meaningful curation, making shopping for clothing on Amazon “a nightmare” according to Bornstein. She recapped a recent Amazon shopping trip:

I was looking for a white bathing suit. I kid you not, there were four pages of the same three items that kept resurfacing. It was unbearable… They were all $14.99. I’m a high- end shopper who is willing to spend more and Amazon should have known that based on my past purchases. But because Amazon earns so much revenue from its ‘promoted products,’ you can’t trust the search results anymore: it makes you wonder ‘Why am I seeing this? Who’s paying for what?’ It’s like Amazon has become the opposite of what we need, unless you know exactly what you want. Sure. If you know exactly what you want and they happen to carry it, it’s amazing…but Amazon is not the answer for fashion.

Bornstein felt confident she could meaningfully improve the user experience by offering tailored recommendations out of the largest fashion catalog much in the same way Spotify customized music playlists from their catalog. An idea began to take shape: a multi-brand personalized online store built around each user, harnessing AI at its core and offering something far superior to what fashion consumers had experienced to date. A critical first step in the realization of this idea was finding a technical co-founder. Bornstein met Aggarwal while he was an entrepreneur-in-residence at Bain Capital Ventures and he proved to be the perfect fit, with extensive knowledge of e-commerce acquired over a 20-year career working on search, personalization, and retail technologies (see Exhibit 1 for team

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI) 521-070

3

bios). Bornstein’s pitch resonated with Aggarwal, who imagined a holistic platform, “One that rethought the business model, the technology stack, and the bringing together of cutting-edge AI with new innovative UX in a seamless experience.”3 The duo spent the next four months getting to know each other and validating their idea. Recalled Aggarwal,

Each of us came at it from different angles. But we came to the same conclusion which is even though people have talked about personalization and leveraging data and machine learning for shopping experiences, the fact of the matter is that things are not that much different today than they were 20 years back. The building blocks of e- commerce haven’t changed. When we go to our favorite website, that experience is not adjusting to our preferences or the feedback that we’re giving to that system. In today’s world, you should have your own store. There’s no reason why you and I have to have the same store in the digital world; technology makes it easy and inexpensive for us to build a million different stores for a million different customers.

The opportunity appeared to be enormous. Fashion was a large market with an increasing share of wallet moving online. Some of this growth was currently stunted by low conversion and high return rates due to the challenges associated with shopping for clothing online. No one retailer dominated in either the physical or digital spaces; rather, retailers competed with one another by offering bigger discounts, more unique product assortments, or additional points of distribution that allowed consumers to shop across an omnichannel system that included both physical stores and e-commerce.

THE YES’s business model proved attractive to venture capital firms, including Forerunner Ventures. Managing partner, Kirsten Green, explained, “Massive market share is up for grabs in retail as transactions shift online, legacy competitors bow, and brands look for Amazon alternatives- particularly in the $300 billion U.S. apparel and accessories market, in which THE YES offers an unmatched, more personalized discovery experience.”4 Noted Tech Crunch, “AI and machine learning already dominate in many verticals, but e-commerce is still open for a player to have meaningful impact.”5 In 2018, THE YES raised $30 million in two rounds from Forerunner Ventures, New Enterprise Associates, True Ventures, and Bain Capital Ventures. Aggarwal explained that this level of funding was necessary to build a completely new technology stack, “Given the complexity of this operation, in the next two years, we’ll have to build multiple technology companies within THE YES.”

Recent Trends in the Apparel and Accessories Industry

The U.S. apparel and accessories industry was worth over $300 billion in 2018 with market share steadily shifting from offline to online, which currently accounted for 22% of sales.6 Given low customer loyalty to both brands and retailers, and fragmented customer acquisition efforts, Bornstein believed there was room for THE YES to stake a competitive claim. In the current marketplace (see Exhibit 2), consumers were confronted with an uncomfortable tradeoff between accessing a broad selection and the ease of finding the right item, and no one seemed to be doing anything about it.

Department stores, historically known for their unique curation of goods and excellent service, had dominated the retail landscape for decades but were declining precipitously in recent years. More fashion brands had opened their own points of distribution both offline and online, which allowed for greater control over branding, merchandising, and promotional activity. Suddenly, a consumer had many more shopping options: she could find a new Cynthia Rowley dress at a department store, at the brand’s flagship store, on the brand’s website, and on ten other multi-brand e-commerce websites. As a result, consumers no longer thought of department stores as the only source for fashionable items.

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

521-070 THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)

4

The shift to e-commerce posed unique challenges for department stores and brands. Weighed down by legacy systems and processes, department stores struggled with digital transformation. Aware of how difficult it would be to set up their own seamless digital experiences, some fashion brands contemplated selling their items on Amazon. However, many, particularly the high end brands, were concerned that this move will erode their carefully crafted brand image. Brands had trust issues and concerns over disintermediation, due to Amazon’s many copy-cat private label brands that it developed after study of a brand’s sales patterns and customer data, and the platform’s unwillingness to shut down rampant counterfeiting. In an effort to fight back against e-commerce retailers like Amazon, some fashion brands and department stores prioritized price promotions over merchandising and service excellence, which further harmed their margins and customer value propositions.

The fall of department stores and the rise of e-commerce led to new phenomena in online fashion (see Exhibit 3): a new breed of multi-branded e-tailers were launched; digitally-native fashion brands emerged; luxury brands, initially hesitant to adopt online channels due to their desire to maintain selective distribution, started their own upscale websites; social platforms, such as Instagram, YouTube, and TikTok allowed brands and influencers to open social-commerce stores. Online competition was heating up.

The Use of AI Applications in Fashion Retailing

Several startups believed helping consumers navigate the endless amount of choice was critical to success in the apparel and accessories market. An industry insider explained, “Psychology teaches us that faced with the overwhelming number of options, we often choose to do nothing. It’s better for retailers to categorize their offerings to help consumers structure their options–and that’s where curatorial platforms come in.”7 Katrina Lake, founder of Stitch Fix, described its value proposition as “we save our clients time by doing the shopping for them,” illustrating the importance and value of merchandise curation based on individual preferences.8

The development and spread of AI was poised to radically change e-tailers’ ability to personalize shopping, and funding for AI companies working in retail tech accelerated significantly in 2019 as more investors were excited by the potential.9 Combined with the vast number of data points that firms could now easily collect from their website visitors, AI-fueled algorithms were ready to allow retailers to build better predictive models that could improve their back-end operations, as well as the front- end shopping experience of their customers.

Two AI technologies, computer vision and natural language processing (NLP), had recently developed that could significantly improve customers’ shopping experiences. “The former helps to index [i.e., tag] products in a website’s virtual catalog using visual cues, while the latter aggregates and learns from words that shoppers use when describing products they are looking for. Both rely on algorithms powered by machine learning, a subset of [AI],” explained an industry consultant. 10 Both technologies organically generated taxonomies by studying pictorial and language patterns across products. These tools were used to improve search results and recommendation and personalization systems (see Exhibit 4). Some e-tailers had been using collaborative filtering to generate product recommendations, which generated item recommendations based on an analysis of what similar customers liked. Others were using computer vision to allow customers to upload images of clothes to enable search for similar or matching designs. The fashion industry had also been an early adopter of conversational assistants (i.e., chatbots) to facilitate a customer’s journey and replicate personalized sales assistance, rather than forcing customers to use the sometimes unwieldy search bar to find their desired items.11

Several fashion retailers had begun to offer tailored recommendations using unique taxonomies and recommendation algorithms. For example, “Farfetch’s product catalogue has more than 3,400

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI) 521-070

5

brands, and its customer base is global, meaning an intelligent taxonomy is important for creating ‘credibility at scale’,” claimed executive Natalie Varma. She continued, “In fashion, there are so many ways of describing the same thing, which is quite nuanced and is quite internal to the industry, but can be a problem for our customers.”12 This imprecision of language made shopping more difficult; as an example, what designer Rick Owens calls a “duvet coat” might be called a “down jacket” at Balenciaga and a “puffer coat” at Prada. The development of taxonomies and the tagging of the thousands of items that populated an e-tailer’s assortment had historically been an arduous and costly process that usually involved human and machine intervention and which needed to be redone and updated each season. For example, garments that arrived at Farfetch were described, photographed, and editorialized manually and then the product’s taxonomy was automatically enriched using a fashion knowledge graph created by Farfetch data scientists working alongside its fashion experts. It stored thousands of descriptive fashion terms that had relationships associated with them, which helped in the product recommendation process.13

While fashion recommendation engines had become more common, the ability of AI to predict new fashion trends was often questioned. Said an industry insider, “Many have claimed to hit personalization right, but it’s hard to do…the real question for personalization is, will this drive better conversion and sales? Furthermore, fashion is fickle; … [with] personalized recommendations based on my preferences, I’m more likely to dress like it’s the 1990s than like it’s the 2020s and that’s not a good thing.”14 Some of the difficulty stemmed from the nature of the fashion industry, where what is “in fashion” is constantly changing, driven by customer tastes, market dynamics, and tastemakers. To address this, some brands focused instead on detecting existing trends and creating new products in real time based on those trends, hoping to capitalize on them before they fell out of fashion. For example, Stitch Fix aggregated all of its customers’ preference data gleaned from their previous purchases and their answers to quizzes to learn which fashion elements were currently popular and then used that insight to design its own private label fashion products within months.15

THE YES’s Algorithm-Driven UX

THE YES wanted to harness the new possibilities offered by advances in AI to challenge and revolutionize online shopping. Their intention was to learn each customer’s own preferences and to use that data to create an individualized shopping feed for each customer that presented only items that she would likely find most attractive. This allowed for curated discovery of products from different brands without scrolling through irrelevant products and/or brands or relying upon knowing enough about what one wanted that one could easily search for it using the search bar.

To learn each user’s unique preferences and taste, THE YES employed a process that included an onboarding quiz to capture general preferences and sizing. This created an initial personalized feed for each customer with a never-ending stream of products that resembled what one would see in a dating app. As a customer scrolled through her personalized feed, she could click “YES” or “NO” to indicate items she liked or disliked, feeding the algorithm with new preference data every time she browsed, which was used to continuously update the feed to improve its predictive ability. Items that were YES- ed were placed onto the users’ YES Lists for future consideration. Users also periodically received optional questions and ‘pop quizzes’ as they shopped that were used to refine their intentions and better understand their preferences. This activity allowed THE YES to capture users’ preference for size, fit, pricing, and brands in a fun, interactive way and to use that data to feed each user’s personalized ML model. Exhibit 5 illustrates the data collection process, and Exhibit 6 provides screenshots from the onboarding process. Bornstein noted, “Each feed is tailored to each user and updates daily and suggests relevant brands, categories, trends, and friends. As the user says ‘yes’ and

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

521-070 THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)

6

‘no’ to items, the data model for each user adapts and learns more about what they like.” The personalization extended also to search functionality, during which an intelligent algorithm used NLP to figure out, based on what she typed into the search bar, what a user was looking for, but also offered individualized search results based on knowledge of their preferences and thus sorted appropriately. Explained Bornstein, “Our goal is not to show you exactly what you told us you like. It’s to show you all the things that we think you might like related to it and help you explore, too.”

Bornstein and Aggarwal decided to initially develop a mobile application (app) for launch rather than a website, noting that 50% of 2019 e-commerce fashion purchases were done via mobile, and that this percentage had been growing steadily. Designing for an app allowed them to build a cleaner UX and to target experienced mobile app customers who were already used to scrolling and swiping to indicate interest or approval on other apps.

Exhibit 7 summarizes THE YES’s offering for which the company had three patents pending for pieces of the underlying proprietary technology. Aggarwal explained, “Building this next generation experience is not an incremental change; it requires a completely different approach. We have built fundamentally new technology in four areas:”

1. Brand integrations: Building a store for every user requires enough inventory to satisfy a diverse set of tastes. This is only possible if we can partner and integrate with brands in a seamless way, one in which there is very little effort on either side. We have built deep technology that can integrate wide and diverse sets of data sources automatically leveraging ML to build a seamless shopping experience for the consumer.

2. Adaptive, user-centric e-commerce platform: Building a user-centric, adaptive e-commerce experience requires a fundamental change in the underlying platform. The platform needs to be AI-enabled, able to read user signals in real-time, support not just text but other types of content such as images, and support the scale of producing millions of stores, one for each customer, instead of just one store for all.

3. Fashion algorithm: We have been hard at work building the best fashion personalization algorithm. This involved fashion experts building the most extensive taxonomy from scratch and then scaling it out through machine learning. Each product in our catalog is now automatically assigned more than 500 attributes and a signature that encodes its style (see Exhibit 8 for an example). Finally, this deep knowledge of products is matched with a small number of high-quality data points from the user to build a ML model for customizing every user’s own store. Unlike traditional personalization that relies solely on aggregated user behavior signals, we use an extensive understanding of style, size, brand, and price to provide the best experience to users (Exhibit 9 illustrates the deep learning model for each user).

4. Algorithm meets UX: We believe one of the key differentiators of our product is how the algorithm interacts with the UX to build a seamless experience. Just like our platform, our UX is built to provide a dynamic experience to the user and one where we are constantly learning. One of the criticisms of personalization has been that it doesn’t allow for discovery. We believe our approach, where we can show new ideas to users and get quick, high-quality feedback from them is a great way to balance personalization and discovery.

THE YES founders took a holistic approach to their customers’ shopping journeys. Explained Aggarwal: “In order to change the customer experience, you can’t just focus on the algorithms, the AI or the ML, you have to think holistically about the user experience and how you bring that technology to the user in a meaningful way and also get their feedback. Customer experience is an end-to-end

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI) 521-070

7

journey so you can’t just focus on one part and make an impact.” Their aspiration was to change the way consumers shop online and to encourage them to change their behavior. For example, instead of needing to search multiple websites for the lowest price, THE YES assured its customers that they would always find the lowest price available on the web, by having the app dynamically track and respond to price changes in real time. THE YES also encouraged consumers to forgo using the sorting and filtering options they used on other e-tailers’ websites. Explained Lisa Green, senior vice president of brand partnerships: “The idea is not to sort from high to low price, but to be able to find exactly what you’re looking for because it’s closest to your style preferences. We know that a shopper often defaults to these functions because there’s no other way to narrow down what they’re seeing. On THE YES, they won’t need to rely on those because the results are already sorted by what they are most likely to buy right on top.”16 Finally, THE YES offered excellent customer service using a high quality, highly empowered team. THE YES relied heavily on human expertise to improve their algorithms. Aggarwal said, “Domain knowledge is super important for building a really personalized experience. Building a great personalized experience across all domains, from fashion to furniture to commodities is not the right approach. You need to start with a specific vertical and focus on it.” Added Bornstein, “Google, Pinterest, and Facebook – they’re all trying to do variations on this, but they’re tackling all categories at once. If you don’t take into account all of the specifics around what actually matters in fashion then it becomes irrelevant and you lost the trust.”17 She elaborated:

[U]nless you understand every dimension about an item and how the consumer thinks about shopping in a certain kind of category, and you understand enough about the consumer within this category, it’s very hard to make good recommendations…to do that, you need humans… Data is only good as the inputs that it gets… if you talk to someone from Pandora… the way that they built their original algorithms is they had musicians listening and identifying all of the elements of music, so that they could then actually build on top of it and understand the relationship between all the elements of music. We did the same thing for fashion. So any given item may have five hundred attributes, and the attributes could range from the length of the sleeve, the color, the construction, the price, the brand, the fit. All of those things are really important to understand as it relates to an item as well as the occasion it could be worn for. Is it dressy? Is it casual? Is it good for spring? Is it good for nighttime?… all of those things matter when you want to build both a search engine and a recommendation engine that are really smart. And so we had a team of human fashion taxonomists building the input, and we continue to add to it, as we’re getting into new seasons and as there are new trends; it never ends in fashion…our fashion director works with a small team to identify all of the trends that are happening…We can surface the right trends for you, and then the algorithm [is] recommending the items within the trend that are most likely to be relevant to you.18

Attracting Brand Partners

Bornstein’s concept hinged on attracting the best and broadest assortment from a variety of brands to compete with the largest fashion retailers both offline and online. To do so, the new platform would need to be as much of a revolutionary solution for brands as it was for shoppers. Most fashion brands used a wholesale model where they sold their products to retailers, who then sold them to consumers. The biggest complaint from brands was a shortage of good wholesale partners in the apparel industry after years of decline in fashion brands’ traditional retail partners, department stores.

Taylor Tomasi Hill, a trendsetter in the fashion world, understood these problems all too well. She joined THE YES as creative and fashion director in 2018 after serving as creative director at Moda

For the exclusive use of J. Li, 2022.

This document is authorized for use only by Jia ye Li in MIS 441 – Global E-Commerce-1 taught by Richard Johnson, Washington State University from Jan 2022 to Jun 2022.

521-070 THE YES: Reimagining the Future of E-Commerce with Artificial Intelligence (AI)

8

Operandi, an online luxury fashion retailer. To

Collepals.com Plagiarism Free Papers

Are you looking for custom essay writing service or even dissertation writing services? Just request for our write my paper service, and we'll match you with the best essay writer in your subject! With an exceptional team of professional academic experts in a wide range of subjects, we can guarantee you an unrivaled quality of custom-written papers.

Get ZERO PLAGIARISM, HUMAN WRITTEN ESSAYS

Why Hire Collepals.com writers to do your paper?

Quality- We are experienced and have access to ample research materials.

We write plagiarism Free Content

Confidential- We never share or sell your personal information to third parties.

Support-Chat with us today! We are always waiting to answer all your questions.