News aggregation is a glory-less pursuit, where even the companies that do succeed have to grapple with the fact that no algorithm is ever going to be as good as the human brain at judging what news/content will appeal to said brain. It’s an AI-hard problem.
This was a mistake. I dismissed the idea initially – not because I am an NLP/machine-learning PhD, but because I’ve watched some of the most capable and brilliant founders struggle to solve the problem of bringing relevant content to you, when you want it, and how you want it: Gabe Rivera of Techmeme, Mike McCue of Flipboard, Kevin Rose of Digg, Garrett Camp of StumbleUpon and Adrian Aoun of Wavii to name a few. In short, it’s a bitch.
Crawling and arranging and parsing links to news articles (or any web content for that matter) is such a pain in the butt that one could just cede the position of the news delivery engine to Twitter and call it a day: Twitter is the Twitter for news, basically.
But that is not what Bradford Cross did. Instead he raised $16.5 million in total funding to bring together some of those aforementioned NLP/Machine Learning PhDs to see if they could find a better way. For Cross, news that you read via Twitter, though somewhat tailored to you through the Follower model, just isn’t relevant enough.
$15 million of that funding is coming from a newly raised Series A round that the company is announcing today. The round was led by Accel’s Jim Breyer with participation by Russian investor Yuri Milner.
The marker of a good news aggregator and discovery service is a combination of familiarity and serendipity. It’s like a good party: If you arrive and you already know everyone, stuff gets really boring really fast; but if you don’t know anyone, there’s no entry point, so what’s the incentive to stay?
Despite my initial rebuff, I started showing an interest in Prismatic again once I started to see more links in my social feeds shared “via Prismatic.” I ended up dusting off the account I had set up back in the day and gave it another whirl, this time actually following interests like The Economist and people like YC partner Gary Tan.
Prismatic’s technology works by crawling Facebook, Twitter and the web (“anything with a URL”) to find news stories. It then uses machine learning to categorize them by Topic and Publication. Prismatic users follow these Topics and Publications, as well as Individuals and the algorithm then uses these preferences and user-activity signals to present a relevant Newsfeed.
Users see this basic feed when they log into Prismatic, and they have the options to view Global News and a Social Feed. The app also suggests users and Topics to follow.
On mobile, Prismatic thus far has hit that magic relevancy + serendipity combination better than anything I’ve seen — showing me not just TechCrunch article after TechCrunch article, but also events like “Kraftwerk the catalogue” going on at the Tate Modern.
I have no idea how it did it, but yes, I am actually interested in Kraftwerk’s catalogue being performed live. Obviously the relevance product, like any other recommendation system, has its dumb moments: For example, I still have no idea why user Morin416 keeps getting suggested to me.
However the more you interact with Prismatic, the more its algorithm learns from you. If you complete actions, such as sharing, saving, commenting or starring stories, Prismatic takes those behaviors as a signal that you’d like to interact with more of those stories. If you drill down to read stories from a given topic, Prismatic remembers that you were at some point interested enough to click and serves up more of those stories. If you select the X button, Prismatic removes a given story from your feed and is less likely to serve you related stories moving forward.
While Cross wouldn’t reveal specific usage stats, it’s rumored that the service has 25K weekly active users (10K logging in through Facebook, according to AppData). What Cross would reveal is that 20 percent of the weekly users return to Prismatic 6 to 7 unique days per week, and that 5 percent of its Daily Actives are sharing stories to Facebook, Twitter and email — which explains all the links I saw.
Like many users I spoke to, I prefer the Prismatic experience on mobile and have no taste for the weekly Top Stories emails, which I promptly unsubscribed to. The elegance of the mobile version is superior to Prismatic on the web, which just went through a redesign but still needs to work out some kinks. The snappiness of the app lends itself well to touch screen: For example, all you have to do to complete actions on iOS is to press down — the share, comment, and save buttons are automatically displayed. Your home Menu screen is a quick swipe away from your news feed.
“It felt like technologically we were at this point where it was possible to do a good job with aggregation,” Prismatic “Chief Software Wench” Jenny Finkel told me last week, when, as penance for being so skeptical, Cross had me meet with five members of the seven-person team in their modest, dorm-esque office in SOMA.
In addition to making a solid cup of cold-brew coffee, Finkel is an NLP/machine learning post-doc who’s made contextual relevancy her life’s mission. “There’s not enough people who know how to put it together,” she continues. If anyone can deal with AI-hard, it’s Finkel and co-founder Aria Haghighi and their cohort of nerd heroes, which investor Jim Breyer called “special.” “Aria and I spent five years staring at text in grad school,” Finkel says very convincingly.
“The ever-increasing amount of online data will require a more sophisticated approach on how we select relevant information,” investor Yuri Milner said when asked about Prismatic’s eventual potential to accurately curate the web. “This new approach could be 50 percent driven by your network and 50 percent driven by algorithms and could make consumption of online data more relevant.”
While the service competes with anything from Zite to Pulse to Flipboard, Cross reveals that what attracted investors the most to the product was the team’s commitment to building a real business. The product roadmap for the next year or so consists of expanding the content types beyond news to stuff like events and eventually apps. The startup does not have “meaningful” revenue presently, but it is conducting some tests with Amazon affiliate links in stories.
This is simpatico with Cross’ broader vision “If people are finding info about an app and then going and buying it and then we want to take them there,” Bradford said, revealing a product roadmap that includes a iPad version natch then Android. He eventually wants Prismatic to become the relevancy engine for all content across all devices. “The right approach is to go slower and be more thoughtful about it.”
Like many an aggregator before it, Prismatic is ambitious and Cross may be crazy. But hey, imagine a world where you could open an app and see a simple, efficient feed of all the news, events and products that were both what you wanted and what you didn’t know you wanted, instead of having to grapple daily with the morass of noise to signal that is the Internet.
Prismatic’s Chief Software architect Jason Wolfe summed it up, “It’s a real giant win for people if we can solve all these problems.”