How Douyin Bootstrapped Its Flywheel

A lot of Douyin’s success has been attributed to ByteDance’s personalization engine, but the personalization engine alone doesn’t explain how Douyin was able to get off the ground. When Douyin first launched its features were so poor that early employees were embarrassed to ask influencers to post in the app. It didn’t make sense to me the the algorithm itself could explain their success, especially early on before Douyin had a lot of organic content to recommend. After leaning into this curiosity I’ve learned that while Douyin’s highly-acclaimed personalization engine contributed to their success, getting the flywheel going was largely due to efforts from their operations team.

The beginning

When ByteDance launched Douyin, the short video market was already hot; Musically had been around for 2.5 years, and Meipai, a leading short video app in China, had 160mm DAU and was making celebrities out of their popular creators. The leader of the pack was Kuaishou which had rural consumers from third tier cities and below on lock.

In order to figure out how to make people like their products in such a crowded market the entire Douyin team conducted extensive competitive research. Team members tried almost all the short video products — about 100 apps — both from China and abroad. Everybody on the team used short video products every day in order to figure out what the strengths and weaknesses of existing players were.

According to ByteDance China CEO Kelly Zhang, the culmination of this research was that the Douyin team realized that none of the products impressed them very much. They decided to focus on four dimensions to differentiate on: full screen and high definition video, music, filters, and personalized recommendations. In effect, Douyin decided to launch a Musically clone. This *should* have failed since Musically already tried launching in China years earlier, and as mentioned the app was embarrassingly bad at launch. Furthermore, Musically was founded by a Chinese team so Douyin didn’t have a local cultural advantage either. The difference that lead to Douyin’s success is the operational effort their team put into getting the app off the ground.

Bootstrapping Culture

Douyin was positioned as a trendy app for fashionable urban elites. In order to ensure that the early adopters had a culture that reflected this positioning, Douyin heavily recruited early adopters by hand. They mobilized their entire staff to lure in creators from competitors, scouting talent on other social networks and short video apps (including DMing Musically creators). To get trendy urban youths creating content, Douyin sent team members into art schools across the country to scout for good-looking students to be its users. Altogether, the team convinced hundreds of them to join by promising to help them get famous online.

After recruiting creators onto the app, Douyin treated them like royalty to get them to stick around. Team members chatted with creators every day, listening to their ideas and making them feel like they were participating in the platform’s early growth and molding its direction. They sent creators custom swag and even sent celebration videos from staff members on their birthday. The best video creators were rewarded with gifts like cameras, celebrity merch, and snacks to make them feel special. One staff member even bought a creator a Christmas tree. These efforts helped to retain creators before user growth was sufficient to retain creators by itself.

The operations team then started to manipulate the visiblity of videos to encourage the type of trendy content they wished to cultivate. Videos that didn’t conform to the community’s tone and values would struggle to gain visibility. This action was especially important given Douyin’s remixing features. Like Musically, Douyin used theme-based challenges to encourage content creation and encouraged users to riff off each other's creations to build upon shared memes. By highlighting the challenges and content they wanted to be remixed, Douyin catalyzed the creation of more content that reinforced the culture they wanted. Of course, users could submit their own challenges and the operations team often collaborated with users on new challenges as well.

Creating lightning

Douyin’s operation team scoured the app for content that had the potential to go viral and amplified it. They set up accounts on other platforms to post watermarked content — including WeChat, which was a big channel for them before WeChat banned them — and went above and beyond to take shots at virality. For instance, when a video mimicking a famous Chinese comedian surfaced on Douyin, the operations team incessantly pinged the comedian on social media until he shared the watermarked video, resulting in millions of views.

Douyin’s ultimate source of power lies in owning consumer attention, but unintuitively the way they achieved that power was by leveraging distribution through existing social channels. This is reminiscent of Instagram being used to post more polished photos to Facebook, and Airbnb auto-posting listings to Craigslist early on.

Why Douyin succeeded when Musically failed

When Musically launched their Mindie competitor they had already spent 90% of their seed funding on a failed education app and had less than $30k remaining. To increase their odds of success Musically launched worldwide which lead to them gaining traction in the US but bombing in China. As Douyin proved, it was possible for Musically to win the Chinese market, but doing so required extensive operational effort that Musically could not afford. It made much more sense for Musically to focus on the US market given the information and funds they had at the time.

Musically eventually re-entered the Chinese market years later, but by then Douyin had already gained traction in China. There was little Musically could to do differentiate since Douyin was already a clone of Musically, had a better recommendation engine, and had a bigger warchest from ByteDance.

The importance of personalization

In TikTok and the Sorting Hat, Eugene Wei insightfully identifies that ByteDance’s personalization engine contributed to Douyin’s crazy high engagement rate and allowed ByteDance to break Musically (later renamed to TikTok) out of the lip syncing niche it was stuck in. However, Kuaishou’s success and even Musically’s success before the ByteDance acquisition indicate that an advanced personalization engine is not required to get off the ground. Musically grew to millions of users with a less sophisticated personalization engine, and Kuaishou is successful (going public at $20bn+ mkt cap) without such sophisticated algorithms — long term Kuaishou users consume up to 50% of their content from creators they follow in-app, whereas TikTok users get 80-90% of their content from new accounts that are popular among other users. Douyin’s operational efforts were a much more important factor in getting off the ground than their more-commonly cited personalization engine.

Additional thoughts

Douyin’s operational playbook is similar to Andy Johns’s technique of establishing a “white-hot coal” of a small but engaged community before scaling. A lot of manual effort is involved in building culture and it’s not a problem that merely throwing money at can solve. In fact, having a lot of money and a large audience may even be an inhibitor since that creates noise that makes it difficult for a specific culture to form.

Bootstrapping culture is very operationally heavy and is distinct from the playbooks of previous social networks which merely relied on a superior product and referrals to succeed. If I were to start a new content community from scratch, here’s the playbook I’d use:

  1. Conduct extensive, global competitive product research before designing the app. Copy the best features
  2. Pick a niche community to focus on first
  3. Recruit content creators from existing competitors and other social/physical networks
  4. Treat them like royalty so they create content until the consumption side is big enough on its own
  5. Manually highlight and reward good content that you want to see more of
  6. Use remixing and shared memes to increase creation of that content
  7. Leverage existing social networks to distribute content and look for opportunities to create virality
  8. Then, feed the flywheel by improving personalization and expanding vertically. Andy Johns has a good framework for how to do this