What is Twitter's (NYSE:TWTR) biggest problem? Slow user growth? Low engagement? Poor ad targeting? These problems all stem from one thing -- the real problem. Twitter doesn't understand what its users want.
For a long time, that worked for Twitter. Organizing things in reverse chronological order provides a decent stream of content. Brand advertising doesn't require the ads to be well-targeted. But it's finally caught up to the company, and it only comes through in numbers like lower engagement, high abandonment rates, and low conversions on direct-response ads.
That may explain Twitter's recent push into artificial intelligence with the acquisitions of Madbits and Whetlab. Now, the company is looking for a couple of engineers to join its Cortex team, focusing on deep-learning artificial intelligence.
The challenge for Cortex
The first task for Cortex's team -- well before it was called Cortex -- was to develop a system to identify unsuitably graphic images posted to Twitter. Twitter, Facebook (NASDAQ:FB), and Google each employ hundreds of humans to help identify images that shouldn't be displayed to the general public without warning. And they're all working on systems that can flag that content. The process is more difficult on Twitter due to the real-time nature of the social network.
Now, Cortex's efforts are focused on improving the company's ad system through deep-learning algorithms, but the details of a recent job listing indicate the company plans to expand well beyond analyzing the content of ads, and targeting them to users. Twitter is looking to hire an architecture engineer to build the backbone of a learning system that can understand the content users are posting.
Twitter describes the challenge thusly: "Twitter is a unique source of real-time information, offering amazing opportunities for automatic content understanding. The format of this content is diverse (tweets, photos, videos, music, hyperlinks, follow graph, ...), the distribution of topics ever-changing (on a weekly, daily, or sometimes hourly basis), and the volume ever-growing; making it very challenging to automatically and continuously expose relevant content."
Twitter is finally focused on the right thing
Those last two words, "relevant content," show that Twitter is finally focused on the things that will get users to engage, and stop them from leaving -- which is one of the biggest culprits in its slow user growth. Early Twitter investor Chris Sacca says nearly 1 billion Twitter users have abandoned accounts.
In 2009, Twitter saw that 75% of users were abandoning their accounts. It found that users following at least 30 people were much less likely to abandon their accounts, and thus focused the product to get users to that number faster.
But this solution created a problem of its own. Now users are abandoning Twitter because the stream of content from 30-plus accounts makes it difficult to find the content they really want.
Exposing relevant content doesn't have to require Twitter to change its timeline -- although it probably wouldn't hurt at this point. It could simply mean better follow recommendations, removing spammy accounts, and optimizing ad placement based on users demographics, behaviors, and the content of preceding tweets.
Facebook is killing Twitter
A recent study from Pew Internet found that the same percentage of users get news from Facebook as Twitter. Twitter's main strength is in breaking news, but the demand for breaking news is significantly outweighed by finding out the most important news. For the latter, Facebook's EdgeRank algorithm, which sorts content for its News Feed, is much better at surfacing the news content users are looking for. Twitter simply relies on trending topics with little individual targeting.
If Twitter really wants to succeed in improving its user growth numbers, its user engagement, and its direct-response ad conversions, it needs to understand the content its users are posting. It can then present the most relevant content to its users instead of making users work to get the most out of Twitter.