Trapit Adapts AI-Driven Personal Search for Paying Customers

When the global artificial intelligence of the 22nd century—let’s call it, oh, Skynet—is writing its autobiography, it will find that some of its grandparents were born in Silicon Valley in the years 2003-2008. That’s when the defense-backed CALO Project was underway at the contract research outfit SRI International in Menlo Park, CA.

Short for Cognitive Assistant that Learns and Organizes, CALO was a $200 million effort to build programs that could reason adaptively and filter information for users based on their needs or their context. The project’s first commercial spinoff was Siri, the virtual personal assistant that’s now part of the operating system in every late-model iPhone and iPad. The second was Trapit, a personalized discovery engine for Web content. Then came Lola, Kuato, and Desti in the areas of banking, games, and travel, respectively—with more spinoffs sure to follow.

Today I want to take another look at Trapit, which has evolved quite a bit since my first profile of the startup back in October 2011. Trapit’s software doesn’t talk back to you, the way Siri does, but it’s arguably much smarter when it comes to finding information that fits your specific interests. If Siri goes down in history as Skynet’s sassy, slightly airheaded grandma, Trapit will be its nerdy, intellectually omnivorous uncle. Its specialty is sorting through disparate data such as news articles on the Web and finding material containing common threads.

A Trapit trap on the World Series

A Trapit trap on the World Series

When the Palo Alto-based startup came out of stealth mode last year, its personalized search service was available only in the form of a desktop-centric website. When the user supplied a topic—say, fly fishing, or vampire novels, or hydraulic fracking—the software would create a “trap” consisting of news articles and blog posts on that topic drawn from 120,000 hand-picked sources. The software was adaptive, observing the user and employing feedback such as thumbs-up or thumbs-down ratings, the time spent reading each article, social media sharing behavior, and other feedback to discern a person’s specific interests with each topic. It would then narrow down the selection of articles it showed in the future.

It was impressive stuff, offering a clear advance over Google Alerts and other persistent, personalized search tools. But I said at the time that Trapit probably wouldn’t become part of my own daily browsing routine until it became more tablet-friendly. The company obliged this July by coming out with a native iPad app—and now the tablet version, which is available free in the iTunes app store, accounts for 84 percent of the time users spend with Trapit.

So it’s pretty clear that building the app was a smart move, at least as a way to pull in more users who could help the startup fine-tune its technology. “What we use the [Web and mobile] apps for is to really hone the algorithms than for anything else,” says Henry “Hank” Nothhaft Jr., the company’s co-founder and chief product officer. “We needed to get people creating and personalizing traps at scale.” Today Trapit has hundreds of thousands of users, Nothhaft says.

But now comes the hard part: locating a few paying customers. Siri, the startup, never had to solve that problem, owing to its rapid acquisition by Apple (which promptly killed the company’s free, standalone iPhone app). But Nothhaft and CEO Gary Griffiths are proceeding as if they’ll need to earn some actual revenue. And it turns out Trapit’s first job will be as a sort of librarian- or curator-for-hire for a big media company, the Malaysian media conglomerate Astro.

Roughly speaking, Astro is the DirecTV, Comcast, and Netflix of Malaysia, with about 3 million pay-TV subscribers, or more than half of the total market in the booming southeast Asian country. The company delivers thousands of local, regional, and international shows to viewers across hundreds of channels. As anyone who’s tried to sort through their cable provider’s electronic programming guide knows, it’s hard to figure out what to watch when there’s so much available. Astro announced in September that it plans to use Trapit to cut through the noise, delivering content recommendations tailored to viewers’ personal interests.

Griffiths and Notthaft say there are more such deals in the works as Trapit matures from a fancy consumer search engine into a white-label discovery platform, available to any company that wants to make it easier for people to navigate large collections of content.

“We have terrific technology under the hood, and our theory was that there were customers who would pay for that differentiation, but they were probably not consumers,” Griffiths told me in October. “I think what you will see over the next few months is not so much us pursuing the consumer under the Trapit brand but other brands who already have an audience using Trapit to provide a better experience for their customers.”

In fact, you can think of Trapit’s Web and mobile apps as prototypes for the tools the startup plans to sell to third parties. Imagine, for example, a rebranded version of Trapit’s iPad app with pre-specified traps that guide users to content owned or curated by a particular company. Such an app might be useful as a marketing tool, a way to repurpose archival content, or a research aid for employees.

A white-label Trapit app for State Farm employees, for instance, might include traps highlighting news about insurance regulation or natural disasters. A Condé Nast app might include traps full of articles from Vanity Fair or Wired. (Those are just examples I made up—Trapit isn’t ready to say which companies it’s actually working with, beyond Astro.)

“This isn’t just a consumer app. The plan all along has been to expose this to business,” says Griffiths. But testing the app first on the broad consumer market was good practice, he says, because it forced the company to beef up its classification algorithms to handle almost any subject. Now the company can start narrowing down its filters so they understand the lexicons of specific markets. “The view was that if we started out very narrow, it would be much more difficult to broaden out later, and we were smart enough to realize there would be all softs of use cases that would emerge that we had never thought of,” Griffiths says. “That said, we never guessed that our first major partner would be a media conglomerate in southeast Asia that would want to use Trapit as a recommendation engine for movies and TV.”

But while the company is putting much of its effort into helping other companies solve their content management problems, it will also continue to refine the free consumer apps, which fill a niche that other browsing tools don’t. “We are often compared to things like Pulse and Flipboard, but we consider those to be the next generation of social RSS readers,” says Nothhaft. “We see Trapit as something quite different from that—it’s really a new way to browse the Web.”

Most of today’s social news readers only show users articles from publications they’ve subscribed to via an RSS aggregator like Google Reader, or links that their friends have shared on Facebook or Twitter. Trapit is broader, pulling in hundreds of articles for each trap from tens of thousands of sites, each of which has been vetted by a human to make sure it’s not a scraping site or content farm. That makes Trapit “unique in terms of its ability to handle a broad range of topics but also bring back high-quality content,” Nothhaft argues.

For real news junkies, the company sees an opportunity to build a subscription-based “prosumer” version of the Trapit tablet app with “more horsepower,” Nothhaft says. “There are features that might be appreciated by only three to five percent of our users, but they might be willing to pay something analogous to what Evernote is charging,” he says. (The premium version of Evernote’s online notekeeping service costs $45 per year.) But there will always be a free, basic version of Trapit, Nothhaft says—if only because “it’s the best form of marketing for the technology that we have.”

As with Siri, consumers may not know or care about the heavy-duty, DoD-funded personalization algorithms under Trapit’s hood. But if the software can find users what they want faster and more reliably, Griffiths and Nothhaft reason, there ought to be parties out there who are willing to underwrite the experience. I’ll check back with them later and let you know how that bet panned out.

Wade Roush is the producer and host of the podcast Soonish and a contributing editor at Xconomy. Follow @soonishpodcast

Trending on Xconomy