The impressive scorecard
Alibaba, the Chinese e-commerce giant, reported a whopping $30.8 billion in gross merchandise volume (GMV) on its online marketplace during the 2018 “Singles Day” – a 24-hour shopping festival. This new record represents a 27% growth year over year, more than doubling total combined sales on Cyber Monday, Black Friday and Thanksgiving Day in the US in 2017.
The beginning of personalization
To sustain and fuel such growth, Alibaba has in recent years invested heavily in machine learning technologies to better equip its e-commerce ecosystem participants – consumers, merchants, payment processors and logistics partners. Some important applications include AI chatbots with real-time translation capability, automated marketing content creation tools, and advanced demand forecasting system to help brands plan “from-idea-to-shelf”.
Arguably the most impactful use of machine learning at Alibaba is the highly personalized Taobao app (its flagship China e-commerce app), a product feature nicknamed 千人千面, or “A Thousand People, A Thousand Faces.” In a nutshell, the backend engine is constantly taking in a myriad of user attributes as inputs, running its proprietary algorithm, and creating a unique and ever-improving app interface for each user. In fact, 千人千面 has become one of Alibaba’s definitive strategies to-date. As users increasingly moved to mobile from desktop (now over 90% transactions happen on the mobile app), 千人千面 solved the critical issue of adapting content from the desktop browser to a much smaller phone screen with more fragmented user traffic.
The results, reported and unreported
To be clear, this is not your everyday app personalization. Alibaba’s almost obsessive data-driven optimization, aided by its expansive reach of user data (the company has full or partial ownership of the leading Chinese apps in ride sharing, mobile payments, food delivery, video browsing, and microblogging), has brought personalization to a whole new level. Since the introduction of 千人千面, app usage and conversion metrics have improved dramatically. Average app-browsing time climbed to an astonishing 45 minutes per user per day.
Indeed, increased efficiency is the key objective of 千人千面. Each square-inch space on the mobile app has become a “monetizable property,” its content carefully curated to ensure high click-through and probability of eventual purchases.
Exhibit 2: Personalized Taobao app homepage
User A: 28-year-old female, active lifestyle, likes food and skincare
User B: 24-year-old female, likes electronics and gaming
Now here is the much overlooked problem. Alibaba knows the user so intimately, that its personalized app is helping users discover brands and products they don’t even know they will like. Oftentimes, users do end up liking these products and become the exact users they were predicted to be. This self-fulfilling effect is even more prominent when the app is the dominant market player, counting 666 million mobile active users in 2018, or 48% of the entire Chinese population. Over time, the yoga-loving woman in her late twenties develops the same taste in snacks as millions of others, while the urban middle-aged man showing a desire for high quality of life (e.g., buys Egyptian cotton bedsheets) becomes an aspiring cyclist with an expensive bike and professional gear – just like all of his male colleagues.
This user convergence and the resulting homogeneity are profoundly alarming. Some top-grossing products emerging in recent years seem just so random: mixed nuts from Costco, “liver detox” supplements from Swisse (a 2nd tier brand in its home market Australia), and 100% cotton wipes from a domestic startup brand. These are likely all high-quality products, but they wouldn’t have reached bestseller status in the absence of the extreme adoption of 千人千面.
The Taobao app’s high efficiency has come at a cost of brand and product diversity. In generating ever-improving conversion metrics, these algorithms pose a severe challenge to the democracy and neutrality of content on Alibaba’s marketplace. Management is aware of the problem, but internal debate continues on whether it should modify this “ruthless” personalization mechanism to give way to more free browsing and advertising.
I believe Alibaba should scale back on the use of 千人千面. In the very least, it can refine the attribute weighting system (e.g., adjust down the weight on previous purchases and searches, and up that on tags like “laid-back” or “liberal”), and designate more app space to algorithm-light content. This may hurt Alibaba’s short-term performances, but will boost its long-term prospects by expanding the universe of products and consumer use cases. As a company with such significant reach and market power, Alibaba should reflect on its social responsibility and closely evaluate the impact brought by its adoption of new technology.
Clearly, the advanced personalization dramatically improves user experience, at least initially. I wonder, is it possible that the algorithm does truly know us better than we know ourselves? Further, is it more patronizing for Alibaba to offer a fully personalized app experience, or to attempt to police the content democracy?
Word count: 793
 TechCrunch, “Alibaba sets new Singles’ Day record with $31B in sales, but growth is slowing”. https://techcrunch.com/2018/11/11/alibaba-singles-day-2018-31b/, accessed Nov 2018
 Quartz, “Forget Black Friday, Singles Day is the world’s biggest shopping event.”
   Form 20-F, Alibaba Group Holdings Ltd, filed on July 27, 2018. https://otp.investis.com/clients/us/alibaba/SEC/sec-outline.aspx?FilingId=12879202&Cik=0001577552&PaperOnly=0&HasOriginal=1, accessed Nov 2018
 2Q FY2019 Quarterly Results Presentation, https://www.alibabagroup.com/en/ir/presentations/pre181102.pdf, accessed Nov 2018