The story behind Toutiao, the $20 billion news aggregator app
Toutiao, a wildly popular news reader app in China, has recently reached a valuation of $20 billion in last round of $2 billion fundraising. Started by a former software engineer Yiming Zhang in 2012, Toutiao experienced exponential growth for the past five years and rode on the waves of mobile internet to gain millions of Chinese users hungry of news and entertainment information.
The secret formula for Toutiao’s growth lies in its focus on data analysis and recommendation algorithms. News aggregator and reader app was always a highly competitive market. Traditional portal websites such as Sina, NetEase, Sohu and Tencent all developed their mobile news app since 2012 and relied heavily on their internal editing crews to provide high quality, curated news services. Toutiao is different from the incumbent in the sense that it does not have the resource to provide manual recommendation and must reply algorithms to understand readers’ interest and preference.
Many newer news aggregator apps fall into the clickbait model BuzzFeed pioneered, full of exaggerating titles and sensational articles. This is because many algorithms tend to only optimize the overall click rate without considering personal preference. The clickbait model is also not sustainable as it can draw too much attention from the Chinese regulator and contradicts with the heavy censorship in the news domain. For this reason, when Toutiao started in 2012, it emphasizes sophisticated algorithm to better understand user taste using data and analytics.
One of its algorithms is called ‘Collaborative Filtering’. It is similar to the clustering or unsupervised learning method in machine learning: users of similar taste are grouped into close clusters and the neighboring user’s preference may affect the ranking of different topics or articles. This requires collecting behavior signals from users such as clicking, favoriting, forwarding or writing comments.
For new users who do not have any behavior data yet, Toutiao intelligently acquired other information to bootstrap the user profile such as analyzing user’s Weibo page and extracting his or her interest label. It also leverages user’s phone model and location to infer its potential income level or spending power.
Toutiao’s effect on data analytics and recommendation algorithms is well paid off. Its daily active user increased from 30 million in 2015 to 78 million in 2016 and now it has over 120 million and the average user time per day hit 76 minutes, making it one of the most addictive and sticky apps in China. As Toutiao dominated the new reader app, it created a fine grain model of Chinese users which became highly valuable to advertisers and brands. In 2017, Toutiao achieved a revenue of $2.5 billion from advertisement, almost tripled the 2016 number of less than $1 billion. In 2018, it is forecasted to reach a revenue between $5 billion to $8 billion in 2018.
Awesome post! The fact that average user time is 76 minutes/day is truly staggering. I wonder how many advertisements a reader sees in that amount of time.
The focus on the value of their recommendation algorithms reminded me of Netflix because their recommendation engine is always raised as a key element of their success. In my opinion, the value of a strong recommendation engine is even more critical for a news company like Toutiao because while there are lots of shows/movies available on Netflix, the amount of potential news stories to read is exponentially greater. And because so much news content is coming out daily and people primarily want to read the latest content, there is no feasible way for readers to identify the ideal stories to read without some analytical help. The other post about Toutiao mentioned that they’re also expanding beyond aggregation to original content – the Netflix similarities continue! Of course, a major difference is that Toutiao seems to be making a killing with advertising revenue, a good use of their personalized customer data (as long as the advertisements don’t get so annoying that utilization decreases).