TikTok and the Sorting Hat — Remains of the Day
English
TikTok and the Sorting Hat August 04, 2020 by Eugene Wei I often describe myself as a cultural determinist, more as a way to differentiate myself from people with other dominant worldviews, though I am not a strict adherent. It’s more that in many situations when people ascribe causal power to something other than culture, I’m immediately suspicious. The 2010’s were a fascinating time to follow the consumer tech industry in China. Though I left Hulu in 2011, I still kept in touch with a lot of the team from our satellite Hulu Beijing office, many of whom scattered out to various Chinese tech companies throughout the past decade. On my last visit to the Hulu Beijing office in 2011, I was skeptical any of the new tech companies out of China would ever crack the U.S. market. It wasn’t just that the U.S. had strong incumbents, or that the Chinese tech companies were still in their infancy. My default hypothesis was that what I call the veil of cultural ignorance was too impenetrable a barrier. That companies from non- WEIRD countries (Joseph Henrich shorthand for Western, Educated, Industrialized, Rich, and Democratic) would struggle to ship into WEIRD cultures. I was even skeptical of the reverse, of U.S. companies competing in China or India. The further the cultural distance between two countries, the more challenging it would be for companies in one to compete in the other. The path towards overcoming that seemed to lie in hiring a local leadership team, or sending someone over from the U.S. who understood the culture of that country inside-out. For the most part, that has held true. China has struggled, for the most part, to make real inroads in the U.S. WeChat tried to make inroads in the U.S. but only really managed to capture Chinese-Americans who used the app to communicate with friends, family, and business colleagues in China. In the other direction, the U.S. hasn’t made a huge dent in China. Obviously, the Great Firewall played a huge role in keeping a lot of U.S. companies out of the Chinese market, but in the few cases where a U.S. company got a crack at the Chinese market, like Uber China, the results were mixed. For this reason, I’ve been fascinated with TikTok. Here in 2020, TikTok is, for many, including myself, the most entertaining short video app going. The U.S. government is considering banning the app as a national security risk, and while that’s the topic du jour for just about everyone right now, I’m much more interested in tracing how it got a foothold in markets outside of China, especially the U.S. with its powerful incumbents. They say you learn the most from failure, and in the same way I learn the most about my mental models from the exceptions. How did an app designed by two guys in Shanghai managed to run circles around U.S. video apps from YouTube to Facebook to Instagram to Snapchat, becoming the most fertile source for meme origination, mutation, and dissemination in a culture so different from the one in which it was built? The answer, I believe, has significant implications for the future of cross-border tech competition, as well as for understanding how product developers achieve product-market-fit. The rise of TikTok updated my thinking. It turns out that in some categories, a machine learning algorithm significantly responsive and accurate can pierce the veil of cultural ignorance . Today, sometimes culture can be abstracted . TikTok's story begins in 2014, in Shanghai. Alex Zhu and Luyu “Louis” Yang had launched an educational short-form video app that hadn’t gotten any traction. They decided to pivot to lip-synch music videos, launching Musical.ly in the U.S. and China. Ironically, the app got more traction across the Pacific Ocean, so they killed their efforts in their home country of China and focused their efforts on their American market. The early user base consisted mostly of American teenage girls. Finally, an app offered users the chance to lip synch to the official version of popular songs and have those videos distributed to an audience for social feedback. That the app got any traction at all was progress. However, it presented Alex, Louis, and their team with a problem. American teen girls were not exactly an audience Alex and Louis really understood. To be fair, most American parents would argue they don't understand their teenage daughters either. During this era where China and the U.S. tech scenes have overlapped, the Chinese market has been largely impenetrable to the U.S. tech companies because of the Great Firewall, both the software instance and the outright bans from the CCP. But in the reverse direction, America has been almost as impenetrable to Chinese companies because of what might be thought of as America’s cultural firewall. Outside of DJI in drones I'd argue one reason DJI had success in America was that drone control interfaces borrow heavily from standard flight control interfaces and are not culturally specific. Thus DJI could lean on its hardware prowess which was formidable. , I can’t think of any Chinese app making real inroads in the U.S. prior to Musical.ly. To build on its early traction, Musical.ly would have to overcome this cultural barrier. It’s been said that if you ask your customers what they want, they’ll ask for a faster horse (attributed to Henry Ford, though that may not be true ). Frankly, that’s always been half horses***, and not just because horses are involved. First of all, what if your customers are horse jockeys? Secondly, while you can’t listen to your customers exclusively, paying attention to them is a dependable way to build a solid SaaS business, and even in the consumer space it provides useful signal. As I’ve written about before, customers may tell you they want a faster horse, and what you should hear is not that you should be injecting your horses with steroids but that your customers find their current mode of transportation, the aforementioned horse, to be too slow a means of getting around. Alex and Louis listened to Musical.ly’s early adopters. The app made feedback channels easy to find, and the American teenage girls using the app every day were more than willing to speak up about what they wanted to ease their video creation. They sent a ton of product requests, helping to inform a product roadmap for the Musical.ly team. That, combined with some clever growth hacks, like allowing watermarked videos to easily be downloaded and distributed via other networks like YouTube, Facebook, and Instagram, helped them achieve hockey-stick inflection among their target market. Still, Musical.ly ran into its invisible asymptote eventually. There are only so many teenage girls in the U.S. When they saturated that market, usage and growth flatlined. It was then that a suitor they had rebuffed previously, the Chinese technology company Bytedance, suddenly looked more attractive, like Professor Bhaer to Jo March at the end of Little Women. In a bit of dramatic irony, Bytedance had cloned Musical.ly in China with an app called Douyin, one that had taken off in China, and now Bytedance was buying the app that inspired it, Musical.ly, an app conceived and built in China but that had failed in China and instead gotten traction in the U.S. View fullsize After the purchase, Bytedance rebranded Musical.ly as TikTok. Still, if that’s all they had done, it’s not clear why the app would’ve broken out of its stalled growth to the stunning extent it has under its new owner. After all, Bytedance paid just $1B for an app that’s rumored to sell now, if the U.S. government approves the transaction, for anywhere from $30 to $70B. Bytedance did two things in particular to jumpstart TikTok’s growth. First, it opened up its wallet and started spending on user acquisition in the U.S. the way wealthy Chinese used to spend on American real estate (no, I’m not still bitter at all the Chinese all-cash offers that trounced me repeatedly when condo-hunting six years ago). TikTok was rumored to have been spending a staggering eight or nine figures a month on advertising. The ubiquity of TikTok ads lent the theory credence. I saw TikTok ads everywhere, on YouTube, Instagram, Twitter, Facebook, and in mobile games TikTok ads are bizarre. The video ads I see for the app in mobile games convey nothing about what the app is or does. One ad I've seen dozens of times has an old lady doing lunges in her living room, another has a kid blow drying his hair, and as he does, his hair changes colors. I feel like the ads could do a better job of selling the app, but what do I know? . If Bytedance could have purchased ads on the back of my eyelids at sub $20 CPMs I don’t doubt they would have done so. It didn’t look like a wise investment at first. Rumors abounded that the 30-day retention of all those new users poured into the top of its funnel was sub 10%. They seemed to be lighting ad dollars on fire. Ultimately, the ROI on that spend would turn the corner, but only because of the second element of their assault on the US market, the most important piece of technology Bytedance introduced to TikTok: the updated For You Page feed algorithm. Bytedance has an absurd proportion of their software engineers focused on their algorithms, more than half at last check. It is known as the algorithm company, first for its breakout algorithmic “news” app Toutiao, then for its Musical.ly clone Douyin, and now for TikTok. Prior to TikTok, I would’ve said YouTube had the strongest exploit algorithm in video, The exploit versus explore conundrum is sort of a classic of algorithmic design, usually mentioned in relation to the multi-armed bandit problem. For the purposes of this discussion, think of it simply as the problem of choosing which videos to show you. An exploit algorithm will give you more of what you like, while an explore algorithm tries to broaden your exposure to more than just what you’ve shown you like. YouTube is often described as an exploit algorithm because it tends to really push more of what you like, and then before you know it, you’re looking at some alt-right video that’s trying to redpill you. but in comparison to TikTok, YouTube’s algorithm feels primitive (the top creators on YouTube have long ago figured out how to game YouTube’s algorithm’s heavy dependence on click-through rates and watch time, one reason so many YouTube videos are lengthening over time, much to my dismay). Before Bytedance bought Musical.ly and rebranded it TikTok, its Musical.ly clone called Douyin was already a sensation in the Chinese market thanks in large part to its effective algorithm. A few years ago, on a visit to Beijing, I caught up with a bunch of former colleagues from Hulu Beijing, and all of them showed me their Douyin feeds. They described the app as frighteningly addictive and the algorithm as eerily perceptive. More than one of them said they had to delete the app off their phone for months at a time because they were losing an hour or two every night just lying in bed watching videos. That same trip, I had coffee with an ex-Hulu developer who now was now a senior exec in the Bytedance engineering organization. Of course, he was tight-lipped about how their algorithm worked, but the scale of their infrastructure dedicated to their algorithms was clear. On my way in and out of this office, just one of several Bytedance spaces all across the city, I gawked at hundreds of workers sitting side by side in row after row in the open floorplan. It resembled what I’d seen at tech giants like Facebook in the U.S., but even denser. The mood was giddy. I could tell he was doing well. He took me and my friends to a Luckin Coffee in their office basement and told us to order drinks off an app on his phone. I reached in my pocket for some RMB to pay for the drinks and he put his hand on my arm to stop me. “Don’t worry, I can afford this,” he said, laughing. He didn’t mean it in a boasting manner, he seemed almost sheepish about how well they were doing. Afterwards, as we waited outside the office in their parking lot, he walked past and asked me if I needed a ride. No, I said, I’d be taking the subway. A Tesla Model X pulled up, the valet hopped out, and he jumped in and drove off. It’s rumored that Bytedance examines more features of videos than other companies. If you like a video featuring video game captures, that is noted. If you like videos featuring puppies, that is noted. Every Douyin feed I examined was distinctive. My friends all noted that after spending only a short amount of time in the app, it had locked onto their palate. That, more than anything else, was the critical upgrade Bytedance applied to Musical.ly to turn it into TikTok. Friends at Bytedance claimed, with some pride, that after they plugged Musical.ly, now TikTok, into Bytedance’s back-end algorithm, they doubled the time spent in the app. I was skeptical until I asked some friends who had some data on the before and after. The step change in the graph was anything but subtle. At the time Musical.ly got renamed TikTok, it was still dominated by teen girls doing lip synch videos. Many U.S. teens at the time described TikTok as “cringey,” usually a kiss of death for networks looking to expand among youths, fickle as they are about what’s cool. Scrolling the app at the time felt like eavesdropping on the theater kids clique from high school. Entertaining, but hardly a mainstream entertainment staple. That’s where the one-two combination of Bytedance’s enormous marketing spend and the power of TikTok’s algorithm came to the rescue. To help a network break out from its early adopter group, you need both to bring lots of new people/subcultures into the app—that’s where the massive marketing spend helps—but also ways to help these disparate groups to 1) find each other quickly and 2) branch off into their own spaces. More than any other feed algorithm I can recall, Bytedance’s short video algorithm fulfilled these two requirements. It is a rapid, hyper-efficient matchmaker. Merely by watching some videos, and without having to follow or friend anyone, you can quickly train TikTok on what you like. In the two sided entertainment network that is TikTok, the algorithm acts as a rapid, efficient market maker, connecting videos with the audiences they’re destined to delight. The algorithm allows this to happen without an explicit follower graph . Just as importantly, by personalizing everyone’s FYP feeds, TikTok helped to keep these distinct subcultures, with their different tastes, separated. One person’s cringe is another person’s pleasure, but figuring out which is which is no small feat. TikTok’s algorithm is the Sorting Hat from the Harry Potter universe. Just as that magical hat sorts students at Hogwarts into the Gryffindor, Hufflepuff, Ravenclaw, and Slytherin houses, TikTok’s algorithm sorts its users into dozens and dozens of subcultures The Sorting Hat is perhaps the most curious plot device from t
中文
翻译失败，保留原文。