Today, TikTok is commonly described as an addictive drug. In a podcast, sociologist Julie Albright described it as:
You’ll just be in this pleasurable dopamine state, carried away. It’s almost hypnotic, you’ll keep watching and watching.
It’s clear that TikTok has captured global attention. As of press time in 2022, they have about 1.6 billion monthly active users. In 2021, its users racked up an estimated 22.6 trillion minutes on the platform — about 43 million years. This was more than double Netflix’s 9.6 trillion minutes in the same year. ByteDance, TikTok’s parent company, makes it look so easy. And with hindsight, it seems obvious that a short-video app would have succeeded the way it did. We can tell a natural story of a progression from text-centric Facebook, to photo-centric Instagram and Snapchat, to video-centric TikTok. But that story of progression is far from what actually happened.
In March 2012, when Zhang Yiming and Liang Rubo started ByteDance in their first office — located in a Beijing apartment — they didn’t start with the idea of a short video app. Instead, Yiming only knew that he wanted to do something to ride the new mobile internet wave. Just one year before, Tencent had released the massively popular chat app WeChat — today, the backbone of the Chinese internet. Yiming was previously the CEO of real-estate search portal website 99Fang — his first CEO role but his fourth stint at a startup. When he started ByteDance, he raised a small amount of money from venture capitalist Joan Wang — who had helped originate 99Fang with him — and two personal friends, for a total of two million yuan of funding.
Yiming and Rubo started out by launching a variety of entertainment apps. Among their suite of apps were “Hilarious Goofy Pics”, “Laugh so much you’ll get pregnant”, and “Real Beauties - Every Day 100 Beautiful Girls”. In The Attention Factory, Yiming looks back on this period:
… one of the unfortunate consequences of building apps with names such as “Laugh so much you’ll get pregnant” and operating out of a retooled apartment office with the CEO wearing a T-Shirt and sandals to work was that recruitment suffered. The most talented engineers they did manage to hire through their own social networks left quickly. “There was no way to persuade high-level vice president candidates to join us. They all thought, ‘How can I possibly partake in something so lowbrow?’ There are even some engineers who won’t join them,” reflected Yiming.
In August 2012, ByteDance decided to launch a new flagship product, something a little more cultured — Toutiao, a news and information app. The goal was to have algorithms recommend content based on past user behaviour, instead of having human editors curate content manually. At the time however, Toutiao’s recommendation system was basic. If a user clicked on an article featuring a female car show model, the recommendation system would guess that the user was probably male. If the user clicked on feel-good articles, then the user was probably a senior citizen. Their clicks along with information like the phone model, time, and location helped build rudimentary user profiles. This was enough to help Toutiao gain more than one million active users by the end of 2012.
It was no accident that Bytedance was so focused on lighter entertainment. Yiming believed it would be more fitting for the new mobile era where time became more fragmented and content consumption became bite-sized. Busy city workers would rarely have the time to sit down and read a 20-minute news article. Instead, they would have three minutes while waiting for the subway to check the news.
Yiming also insisted on improving the recommendation engine for Toutiao, believing that it would be critical for the company’s future. He scoured for leading edge books, connected with researchers, and acquired outside talent. He poached Baidu’s staff like their Deputy Director of Search, Yang Zhenyuan, in 2014 and continued to hire other recommendation-system talent for the next few years.
While Toutiao was improving its recommendation system, ByteDance found ways to grow the popularity of Toutiao. It engaged a grey market network of phone distributors to pre-install the app on cheap Android phones. When that new phone user turned on their phone, they would see Toutiao pre-installed and maybe give it a try. If they did, the recommendation engine would start working and track everything that they clicked on. So the next time they opened Toutiao, the app would serve up personalised content which lured them to stay longer. The more time spent in the app, the more personalised the content became. This virtuous cycle for Toutiao helped them gain tens of millions of users.
By 2015, when the ByteDance team started to discuss their future direction at a company retreat, Yiming brought up the idea of trying short videos. The idea wasn’t new. China’s tech giants were already aggressively promoting their short-video apps — the country’s subway stations were plastered with ads. Tencent had Weishi and Weibo had Miaopai. Other apps like Meipai and Kuaishou seemed to be early winners of this new market. ByteDance might already be too late.
But the data was too compelling to ignore. Toutiao, which included short videos, experienced impressive growth in usage. In Q1 of 2016, the average active user spent 53 minutes on Toutiao. By Q3, it had increased to 76 minutes with half of the growth coming from short videos alone.
ByteDance entered the nascent short-video app market in 2016 by copying from the best. They would launch three apps. “Xigua Video” imitated YouTube. “Huoshan Short Video” imitated Kuaishou, China’s leading short-video app. And “A.me” imitated Musical.ly, the Western market leader of short videos. Of the three, ByteDance had the least confidence in A.me. They had already witnessed the failure of Musical.ly in China and its growth was plateauing in other international markets. Another indicator was the brief success and failure of Xiaokaxiu in 2015, a Chinese version of Musical.ly. Why would A.me be any different?
The initial data gathering also looked unpromising for A.me. When the team of ten engineers asked nearby middle school students what kind of apps they liked to use and why, the students could only make judgments based on the apps that already existed. It was difficult for them to imagine enjoying something new like A.me. It reminded the team of the quote from Steve Jobs: “A lot of times, people don’t know what they want until you show it to them.”
When A.me was launched in September 2016, there was no explosive growth or media fanfare. Their initial user base was so low that when they decided to remove staff accounts to get a more accurate picture of the user base, the active users stats dropped by half. One employee recalled it as a “bleak time” and said that she “didn’t know what to do”.
The team moved to enlist established creators from other platforms to create videos on A.me. The thinking went that those established creators would inspire other early users to create videos themselves too.
This small group of creators were treated like royalty. The operations team would chat with them individually to listen to their ideas and make them feel like they were helping mould the platform. Some in Beijing were invited to free meals at ByteDance’s company cafeteria. Other top creators were featured in the app and given gifts like cameras and snacks. The A.me team even went so far as assigning account managers to individual creators who would occasionally buy them dinners, help with school assignments or relationship issues.
But it wasn’t enough. For the first half-year after launch, app performance was poor. The A.me team half-expected ByteDance to kill their project. Yiming, however, reasoned that it was worth continuing. He believed that A.me was on the right path, just that they hadn’t done a good enough job executing. Matthew Brennan writes, in The Attention Factory:
Given the poor performance, no one would have thought it strange had the team been shut down. “To abandon a project, mainly we look at the data,” described ByteDance executive Chen Lin, “but the leader also has to use their judgment.” A leader needs to judge why the data is bad.
Is the market segment smaller than they thought? Have they misjudged people’s needs? Or did they just execute poorly? Perhaps ByteDance doesn’t have the right company DNA to do something targeted at such young people? All these were legitimate doubts during these early months.
In the end, Yiming reasoned as such: “The logically correct thing is definitely right. And others have already verified (this path), our data is poor because we haven’t done a good job ourselves. (emphasis added)”
The team continued to fix bugs, to improve the app’s features (which was criticised as too basic), and eventually executed a full redesign. The unintuitive app name of “A.me” was changed to “Douyin”. Dou (抖) can be translated into “shake” and yin (音) into “music”. The new name came with a new logo, a musical note shaped into the letter “D” (for Douyin) with a glitch effect on top of it. The intended effect was to be a brand that could tap into the target demographic of trendy, urban youth.
With the visual redesign complete, the Douyin team went deep into China’s art schools and recruited good-looking students to be its users. They got several hundred of them to join in order to reposition Douyin as a cool and fashionable app. The operations team contributed to the repositioning by adjusting the visibility of certain videos — promoting on-brand, trendier content in order to shape the creative direction of the app.
ByteDance then went on a spending spree. It advertised the refreshed Douyin with playful advertisements in cinemas and online — ads that quickly went viral. They also sponsored a new talent show, The Rap of China, which was specifically designed to appeal to the country’s urban youth. The show quickly became popular with Douyin’s intended target audience. Finally, ByteDance hosted a promotional event to celebrate the app’s first anniversary.
China has a week-long national holiday every year, in October. In 2017, over the course of that one critical month, Douyin’s daily users doubled from seven to 14 million. By the end of the year, they had reached 30 million daily users. At the same time, the 30-day retention rate improved from eight percent to over 20 percent, and in-app time doubled from 20 minutes to 40 minutes. These improvements were largely attributed to Zhu Wenjia, a leading algorithm specialist from Baidu. He was hired by ByteDance in 2015 and was assigned to work on Douyin in September of 2017. His team developed Douyin’s original content recommendation engine — a machine learning system that would play a big role in ByteDance’s future growth. Brennan writes:
The better the metrics, the more resources ByteDance placed behind the app as it now had good retention and was fast-tracked into becoming a strategically important product. Suddenly support was coming in from all over the company—people, money, user traffic, celebrity endorsements, brand collaborations, and most importantly, full integration and optimization of ByteDance’s powerful recommendation engine. Chinese stars with massive fan bases such as Yang Mi, Lu Han, Kris Wu, and Angelababy opened accounts, joining in publicity campaigns, and a nationwide “Douyin Party” event roadshow was planned. Douyin had become the hottest upcoming app in China.
In 2018, Douyin had a daily promotion budget of US$2.8 million or CNY 20 million. As ByteDance continued to improve their content recommendation software and spend big on advertising, they began to reposition Douyin again.
Initially, A.me's target audience ranged from teen girls to those in their early twenties. A year later, in 2017, Douyin sought to become the app for urban youth to showcase their fashion, dance, and music. In 2018, ByteDance deliberately changed their promotion strategy once again to invite a wider group of users. Travel, food, sports, games, pets, and all other niches were welcome on Douyin. Their slogan changed from “Let worship start here — devoted to the new generation of music short video communities” to a much more general “Record beautiful life”. By the end of 2018, Douyin had more than 200 million daily active users.
Attention Factory by Matthew Brennan. Published in 2020.