With news coming out seemingly every other day about the drama of TikTok I thought it would be appropriate to discuss it. This will likely be a series as there’s so much to discuss about TikTok, today I’ll just be focusing on the recommendation algorithm.
Trump gave ByteDance a deadline of September 15 to finalise a deal with a US company to sell its US (and likely Canadian, Australian and New Zealand) operations in order to protect user data from the Chinese government.
ByteDance has been in talks with several potential US buyers including Walmart, Microsoft, Oracle and Triller (a TikTok competitor) since the threat last month.
China is now opposing the sale of TikTok’s US operations citing national security reasons, saying it won’t allow the export of confidential AI based systems such as TikTok’s algorithm. Beijing is adopting a hard stance on this now saying they would prefer to see the app shut down rather than be sold.
Now that you’re all caught up let’s talk about the thing that is causing all this drama. The algorithm.
The Star System:
TikTok’s algorithm owes its origins to the Hollywood star system where in the early days of entertainment Hollywood execs would select talent not on their ability to act but on their ability to market that talent and turn them into stars.
TikTok’s algorithm algorithm was designed to be like this from the very beginning.
“Building a community from scratch is like discovering a new land, in the beginning you have to build a centralised economy meaning that from a wealth distribution point of view you make sure the majority of the wealth is distributed to a small percent of people.” Alex Zhu - Co-founder, Tik Tok/Musically
The algorithm was designed to build a community. Alex Zhu, in a rare keynote with Greylock Partners reveals his unique (and honestly amazing) view on building a community. He compares building a community for TikTok to running a country/economy.
If you imagine TikTok as America, a new undiscovered land then Facebook & Instagram are Europe. Europe already has a developed economy and population so the question becomes how do you attract people to move to America?
The problem with Europe is that the social class is already stabilised. The average citizen has almost zero opportunity to elevate their social class. So, in the beginning you have to build a new centralised economy, meaning that from a wealth distribution point of view you make sure the majority of the wealth goes to a very small portion of people to make sure these people get rich.
TikTok handpicked early adopters to skyrocket to fame much like the Hollywood Star System to act as the role models/case studies that inspire the majority. They see this normal person go from obscurity to stardom in America and so people in Europe start to believe this is attainable. This is how TikTok first started building a critical mass.
We see this with the likes of Charli D’Amelio and Addison Rae. Charli, the original poster girl for TikTok amassed over 35 million followers in less than a year and currently stands at 86.2 million at the time of writing. A second-order benefit was all the free earned media TikTok got due to these influencers’ meteoric rises (e.g. Charli D’Amelio on Late Night).
Next, you have to decentralise the economy. “Having an American dream is good but if it’s only a dream people will wake up eventually.” If people start to realise after a few weeks/months this is a dream/unattainble people will quickly lose motivation and hope. So TikTok decentralised their traffic model and gave opportunity to average people to build some traction. This is TikTok’s way of building a middle class. It’s so fascinating to see the comparisons between China, who are currently growing their middle class and TikTok’s (a Chinese company) strategy. It’s my opinion their massive growth was only possible due to Alex Zhu’s unique insights as a result of his Chinese origin.
The importance of a strong middle class cannot be overstated. There are countless examples in the “real world” where countries and regimes have collapsed in part due to inequitable wealth distribution. But perhaps more relevant to this conversation, it’s why Mixer failed (I may explain this is a later article).
TikTok specialises in keeping the dream alive for these new creators and it’s what makes them so special. Creating content on TikTok feels like a gamble, you never know how many views you’re going to get but they let you know if you upload consistently you’ll eventually find viewership and likely going mini-viral.
Next is the element of control the team over at ByteDance have. In order to reduce churn TikTok wants you to have the best and most entertaining experience. Sounds obvious right?
What’s not so obvious is that the team decided to implement a list of content curated by humans to sit at the top of new users’ feeds to impress them with the quality of content on the app.
This allows them to include a variety of content deemed high quality to better determine what the individual enjoys and better tailor content moving forward whilst making the user associate TikTok with an incredibly high base level of content (a clear differentiating factor for a UGC platform).
Content Creation Tools:
Initially Zhu’s team chose to focus on the utility aspect. Their rationale was that the very first users weren’t using the app for the likes but for the amazing features that weren’t available elsewhere. Instagram 1.0 was attractive because it allowed you to host your photos in an aesthetically pleasing, clean way - people weren’t looking to become influencers. This focus on utility before building a critical mass led them to building their content creation tools allowing users to edit from their phone and utilise effects such as slow mo, filters etc. without leaving the app.
So you’ve now got a platform with an incredibly sophisticated star system paired with an incredibly low barrier to entry through their content creation tools combining for a very healthy middle class. And then comes the magic of the for ‘For You’ page.
This is where it becomes important to understand both the history and current portfolio of ByteDance, TikTok’s owner.
ByteDance’s first breakthrough product was TouTiao, a news app that focused on the feed and the algorithm from its inception. TouTiao would use every piece of user information, from which stories users skipped to how long they choose to read them (engagement rates) to feed back into a feedback loop refining what the user would see to personalise their experience. TikTok’s basis of building from TouTiao differentiated them from typical social network platforms and allowed them to revolutionise social media.
The power of TikTok’s algorithm also comes from ByteDance’s portfolio:
They’re able to best test new features on their other platforms to make more informed choices
They’re similarly able to pull user data and proprietary IP from other properties to improve their own algorithms
TikTok is built to be a discovery platform not a social media site. It’s this key difference that makes people start creating content and believing they can find fame/attention on the app. When a user opens up TikTok they are immediately brought to the ‘For You’ page. This is a continuous stream of content that isn’t just serving you content from your friends, but content their algorithm believes you will find most entertaining. The first benefit is that this means your ‘For You’ Page is infinite (no cap on daily active engagement).
Second, what turns so many other content creators off starting YouTube/Twitch channels or Instagram pages is the high saturation, where it feels impossible to ever break through the noise and get noticed. As many TikTok creators will attest to, if they are consistent one video will likely blow up and get 100x the viewership of a normal video. This mini-virality is incredibly addictive for creators and immediately gets them hooked on TikTok content creation.
Their final advantage over YouTube/Twitch is their content form. Because TikTok is for 15-60 second short form videos that they choose to show you TikTok is able to show you hundreds of new pieces of content every hour. In YouTube’s case videos are on average 10-15 minutes, meaning the frequency they can show you is far lower and the majority of viewership comes from subscriptions (doesn’t give much additional viewing data).
This focus on discovery is so valuable because once you get users creating content there are a few second-order effects:
Churn goes way down
Increases number of daily active users
Increases average time spent on the app
The Looping Mechanism:
From a nerochemical perspective, the loop is brilliant. It serves to ensure that commonly used audio clips such as '“Don’t be suspicious” get stuck in people’s heads for days meaning TikTok is always front of mind.
However, I think the loop is one of TikTok’s very few weaknesses. There comes a point where you may be washing be reaching for something or get distracted and hearing the same 15 second audio clip becomes infuriating rather than funny.
This problem is easily fixable. If TikTok were to give us an ability to set a maximum limit on plays so that the feed can then move on autonomously, it would reduce the annoyingness whilst still allowing users to scroll back up if they missed something.
This video wouldn’t have been possible without Alex Zhu’s keynote at Greylock Partners and Colin & Samir’s YouTube video “TikTok’s Takeover, Explained”. Both are fascinating and I highly recommend checking both out.