Automated Serendipity

I don’t know why college kids are especially enamored of StumbleUpon, the content recommendation and discovery engine. I have been covering this curious company for a while, so I was surprised when my then-19-year-old daughter told me it was her favorite app.

The company has accrued 36 million registered users and millions of app downloads since it offered a mobile version years ago.

“Our sweet spot is 18 to 34, but the key demo are college students, 18 to 24,” says Amol Sogal, product manager, mobile.

Go figure. Sogal says that the company sees people tweeting all the time about curious use cases for the discovery apps. One student recently shared how he discovered a volunteer program overseas via StumbleUpon. In stark contrast to other online experiences, “you don’t have to scroll through your feed,” adds Sogal. And unlike general searches that front load the obvious results, “we get you to that third or fourth page you might never have gotten to.”  



The most fascinating thing about StumbleUpon is the range of signals it is using to personalize recommendations. Beyond just age and gender and declared interests, it is working in Pandora-like fashion, registering consumption patterns and likes and dislikes. But it is also using day parts to calculate what content might be most relevant at both a given time and place. “We take a data science approach and look at behavior around content,” says Sogal.

StumbleUpon puts a premium on delivery speed because mobile is a medium based on rapid-fire browsing. So a good deal of work goes into loading content in background as the user is deciding whether to drill into the current piece. And the system is constantly being honed with samples and testing. “When we get content, we sample it out to the user base and can infer how well the content will perform,” he says.

Users declares some interests to get the personalization started. Then they swipe through small cards with thumbnail previews of the story, while in background a fuller rendering of the page loads.

The company uses the same technology to drive the native ad model. As you flip through content cards you are likely to come upon a highly personalized and targeted page that is sponsored. But Sogal says most users are barely aware it was an ad at all.

The next stage for StumbleUpon is to enhance its role in creating communities around the content. In the relaunched Android version of the app, an Activities Center lets users access all functions such as shares and alerts from other users. In the end, cultivating this social layer helps make the machine even brainier. “By connecting you with the people you know, we can make a lot of inferences about the content you like,” he says.  

StumbleUpon seems to endure partly because it is a nice counterpoint to much of the rest of the Web experience. In the  search-driven, task-oriented realm, we are used to machines responding to our specific needs. StumbleUpon’s  algorithms are aimed at uncovering the things you may not have known you are looking for.

There is a lesson here. Even though the product is as mechanized as it gets -- driven by data and optimized on the fly -- its end result is to make users feel as if they are interacting with something more human.

“It is serendipity with the press of a button,” Sogal says.  

1 comment about "Automated Serendipity".
Check to receive email when comments are posted.
  1. Sadie Marshall from O&S Media, February 24, 2015 at 3:44 p.m.

    Great article, really enjoyed reading this.

Next story loading loading..