A Close Look into Tik Tok
Tik Tok, or Douyin (抖音) is a short video social app, and the MOST downloaded app worldwide on App Store in Q1 2018[1]. Launched in Q3 2016, it has reached 150M MAU, DAU / MAU as 45% and 50.5% users will spend more than half an hour on the app per day. 93% users are under 30 [2]. Outside China, in Japan and Thailand, it has reached #1 in free app category.
I spent some time collecting reports from difference sources and played with the app. Here’s my understanding of its mechanism and what we could learn from it.
tl;dr:
- Video content delivery is by recommendation system, based on a mix of manual curation and machine ranking. Follow feature is de-prioritized. Platform has more control to set community trend and distribute exposure across long tail creators. New users also don’t need to set up by finding accounts to follow.
- Mimicry is the biggest driver for normal users to create content. Video templates / concepts developed by power users or hired creators lower the barrier for content creation. Users learn concepts via browsing discovery feed.
How it works in a nutshell
The main interface of the app is a discovery feed auto playing video in full screen. Video is 5–15s long and with background music. Swipe down to see the next one. It also has search page with hashtag challenges and follows page but the majority of viewers stay in discovery feed. The app also come with an in-app composer with recording tools (i.e fast forward / slow motion) and filters.
How the videos on the platform look like?
Most of the videos on the platform are funny videos. Based on survey[2], 82.3% of users indicated they go the the app to see funny videos. The following categories are talent (56.0%) and daily life (54.1%).
Music sets the concept of the video
When posting a video to TikTok, you must select a music clip. Traditionally music is an optional background enhancement for video. However on Tik Tok, music is the soul. It set the theme and acts as a script / template on how the 15s story should be composed. Here are some examples:
That’s one of the most popular video on TikTok. The song is an edited version of Tylor Swift’s <Trouble>. The song is injected with three screams to create a concept. Videos using this music should have turning points at the right moments of the screams. In this example, the bill the woman held got smaller and smaller upon each scream and finally became just coins.
Here’s another example:
The music’s theme changed at 10s, so video story should have a turning point there. In this example, the cover on the kid’s profile photo moved away, changing the photo from serious to funny.
Usually the songs popular on TikTok are less about the popular songs in billboard but more funny and dramatic.
Each video is 5 seconds to 15 seconds short and feels super lightweight to watch. It fits into the scenario where users just want to check their phones and have a few minutes to kill.
We should have seen this type of music video already back to 2013. During that year, Harlem Shake was extremely popular in United States. The idea is basically the same but in Tik Tok, there are hundreds of different music video concepts like Harlam Shake.
Content Creation
Almost all videos on the platform are user generated. Tik Tok did a great job in getting normal users creating videos.
Mimicry is the key motivation for content creation
Based on user study[2], 77.8% of users indicated their motivation of video creation was wanting to try by themselves when seeing funny videos. Its noticeable that only 16% indicated they want to get followers.
Each popular song has several templates for user to follow. Tik Tok has music page where it aggregates all videos using the same song. You can see those videos using the same song can be categorized into several concepts / templates. Here’s the example of a hysteric song. The most common concept here is holding a hairdryer and lipsync the song:
And under the song’s page, we can see half of the top videos are using this concept:
Users may not just copy the concept, they will also contribute to that concept community by adding their own little creativity. Here’s an example of a pretty awesome ‘mind move’ concept:
And later on after many similar videos, here comes a new version:
For users who create content, the normal flow is usually browsing through the discovery feed to watch funny videos. If they stumble upon an interesting one which they feel they can mimic, they will save it and mimic later. On Tik Tok, creating a video is still very time consuming. On average it takes more than an hour for user to create a 5–15s videos. However given the amount of UGC videos created on the platform, we may learn that to user, the key blocker for creating video content is the idea or concept. With the right concept inspiring users, they are willing to spend reasonable amount of time creating content. The platform has mechanism to intentionally promote more mimic-able videos to the discovery feed which we will talk more later. It’s also reported that Tik Tok’s operation team hires creators to create these concepts.
Toolkit may not be the key
In the built-in video recorder, Tik Tok does provide rich variety of toolkit including recording mode (fast / slow motion) and filters / masks. However, browsing through the discover feed and checking the the top videos under each song, you would only see not many of them using fancy effects. Quite a few use the recording tools like slow motion / fast motion (many dancing videos are using fast motion to make the dance looks smoother) or record several clips to make a video but that’s it. More videos are based on concept level effects instead of built-in tools. One example is the bottle effect which is pretty popular:
Another great example will be the doge mask. Like all other apps (Facebook, Snapchat, FaceU), Tik Tok also has quite a few AR face masks. But it’s really interesting to see users in Tik Tok not only simply put on the mask, but also create a very viral concept based on the mask:
Basically in this concept, a group of people will dance with the music and the one with the mask on will turn their face or cover it so the mask will be applied to the next random people because facial recognition can no longer recognize the previous face. The mask itself has nothing special but it is the concept that makes the video fun and motivating for mimic-ing.
After creating such videos, users also tend to reshare it on major social networks such as WeChat or Weibo. Based on the study [2], 56% new users download the app because they see their friends resharing their videos on other major social platforms (WeChat / Weibo). When sharing on other platforms, the video has Tik Tok’s watermark on it. My personal hyphothesis is for quite a few normal users, on content creation side, they see Tik Tok more as ‘concepts pool’. They create contents majorly to show off among their friends and they don’t care much about the distribution on TikTok.
On the other side, the platform also holds hashtag challenge campaigns such as #myrobotdancemove. It worths to mention that for each hashtag, the platform will provide the initial demo video for users to mimic. Several popular concepts are initially introduced via challenges. This is more like promotion for specific concepts.
Content Consumption
In previous part we talked about how important the discovery feed is to surface interesting contents to users, let’s talk more about it on viewer side here.
The discovery feed is the first screen and is where most users spend the majority of in-app time, similar to News Feed to Facebook. However, unlike Facebook’s News Feed, the content on discovery feed is fully controlled by the platform and the followed accounts page is the second tab. Only 22.5% users in user study [2] said they have ever been to the followed accounts page. This design has several benefits:
- Cold start is much easier. New users don’t need to find accounts to follow to start enjoying all the contents.
- Platform has more control on setting the trend / community style by deciding what type of content to promote. So the platform actively promote contents that feel more align with its branding (young, cool, fun, fashion) and have potential to go viral in mimicing.
- From long tail content creators’ perspective, platform controlled distribution provides them an opportunity to get tremendous exposure without building a huge fan base, which is extremely hard and costly nowadays on major social platforms like Facebook (WeChat) or Twitter (Weibo). As long as the content itself is creative, there is a chance that it will be picked by the platform and go viral. It’s more like a fair game for everyone at the same starting point. Sometimes the likes one celebrity’s video received are less than a funny video by nobody shot on their phone.
How they implemented the discovery feed is via a mix of manual curation and machine learning. Initially a new video is uploaded to the platform, the operation team will review it. The team will not only check integrity side to make sure it doesn’t violate community standard (violence, nudity etc.) but also see whether the content is creative and has potential to go viral. If so, press a button to boost it to get more exposure for cold start. After the boost, machine learning will determine whether it should go further by checking metrics like play through rate. Because of the first manual pass, it’s guaranteed more stylish and creative contents could get chance of getting more exposure, while ‘stylish’ and ‘creative’ are hard metrics for machine learning to measure at beginning.
Another thing to mention is the user interaction in discovery feed is very simple: every video is played in full screen automatically. The only user interaction needed is to swipe up to the next one. Its much optimized passive watching experience. Many users reported they love the design because its immersive and they don’t need to make any choice (compare with Youtube or Watch). It fits the scenario where they just have a minute or two and want to find something interesting to kill the time.
Challenge TikTok faces
As we can see, TikTok highly rely on the ecosystem where concepts in discovery feed trigger mimicry, resulting more contents into feed. This could result with many similar contents on the platform. Right now users still enjoy these contents since users will add variance to the concept (micro creativity as mentioned), however, whether at some point users will get bored of the concept is unclear. Based on the survey [2], 48.0% churned users chose ‘too many similar contents / get bored with the content’ as reason. Whether it could continuously find new creative concepts with enough diversity will be a key for its long term success. However, as it has already acquired that many users and # of uploaded videos is high, chances are there should always be some great new videos among them.
Conclusion
One lesson I learnt from TikTok is content discovery can really help content creation. When building products, we tend to focus much more on what kind of toolkit to build but less on how to really continuously teach and inspire users how to use it. TikTok’s team, especially the operation team, did a great job in creating / finding inspiring video contents and surface them to every of its users.
Reference:
- The Top Mobile Apps, Games, and Publishers of Q1 2018: Sensor Tower’s Data Digest, SensorTower, https://sensortower.com/blog/top-apps-games-publishers-q1-2018
- User study of TikTok and Kwai, Tencent Market Insight, https://xw.qq.com/tech/20180409002763/TEC2018040900276300/
