As you grow older and busier, it becomes more difficult to make spontaneous discoveries. Or at least that’s the theory behind a bevy of so-called predictive apps purporting to know each user well enough to hand them their next favorite song, restaurant, or magazine article.
I gave these tools a test run on a recent trip to D.C. With my best friend from college, who is now a chef. It was my first visit to D.C. Since the eighth grade; I knew nothing about the city and needed all the help I could get. I left the laptop at home and went strictly mobile for two days, bringing only my iPhone and iPad. (We also tried Airbnb for the first time.)
For restaurant ideas, I turned to Ness. Corey Reese, 27, co-founded Ness Computing in 2009. The app produces a “likeness score,” a percentage that denotes how likely you are to like a particular recommendation. Reese says that Ness could eventually become a personalized search engine, but for now the company is focusing on restaurants and cafes. He brags that users keep telling him, “It feels like Ness knows me.” Mark Johnson, the 34-year-old CEO of newsreader Zite, says the same thing. Zite users, he says, rave: “My Zite knows me.” It should come as no surprise that more than a few smart people in tech are working on recommendation engines. “We think your entry point for ‘Where should I eat with my friends’ or ‘What’s the cool store nearby’ is happening on mobile now,” says Reese. “It’s happening when you’re already out and about.”
How does it work? When you first open Ness it asks you to rate, on a five-star scale, 10 restaurants near your current location. I did this in Manhattan before heading to D.C. And I found the process flawed. Because the app doesn’t distinguish between levels of cuisine, it will ask you to rate Daniel Boulud‘s pricey DBGB on the same screen as it asks you to rate a local Burger King (BKW). It also includes Starbucks (SBUX). Similarly, it asks you to rate bars that happen to serve food. Sure, I like Brother Jimmy’s — for playing beer pong. Their wings are okay, but if I give it four stars, will Ness start offering me frat bars exclusively?
Once I was in D.C., Ness fared better. Based on my NYC ratings, it offered us a Mediterranean place, some Middle Eastern fare, and a few upscale American restaurants my chef friend already knew about. Ness includes, along with its percentage prediction, a price rating so you know what you’re getting into. It’ll also tell you if a place is the first, second, or third most popular restaurant of its type in the city. (Zaytinya, which it offered up, was the most popular Mediterranean in Washington.) We chose Central Michel Richard (Ness promised an 82%) and enjoyed the meal.
After lunch, connected to Wi-Fi in the apartment we had rented on Airbnb, I spent some time with Zite, a so-called “intelligent magazine” a la Flipboard. Though Flipboard has been the buzzier of the two, Zite seems to learn its user’s reading habits better than Flipboard, even if you choose not to connect it to your social media. I had been using Zite for a few weeks and, indeed, found that the more stories and articles to which I gave thumbs up or down, the better it was getting with the stories it displayed on my personalized front page.
Zite allows you to add sections, giving it a newspaper feel, and although my own are rather generic (sports, gadgets, literature, fiction, marketing), they can get more hyper-specific than you’d ever think of wanting (knitting, mixology, quantum physics, and consciousness are all options). It is not without its flaws: The app once gave me a sports story from Oregon’s McMinnville News-Register about two NCAA football teams I care nothing about, and in the Boston section I’ve added, stories frequently appear that have nothing to do with Boston apart from a quick mention. But its errors are rare and become more so the more I use it. When I fired it up in the apartment that day, I was genuinely interested in four of the five articles that showed up on page one. In total I spent nearly an hour swiping through the 19 pages of my main section, reading articles both short and long. Not bad.
Johnson, the brainy geek behind Zite, is a Stanford Philosophy major who spent three years at SAP (SAP) and then worked on Microsoft’s (MSFT) Bing. He also worked for search startups like Powerset and SideStep. Johnson is enraptured with the idea of perfecting a recommendation engine, which, after all, is what Bing is supposed to be. Such a thing, really, could never be perfect every time, nor would you want it to be: From a developer standpoint, he says, “the problem with any of this stuff is that serendipity is pretty easy to create — it’s what we like to refer to as randomness — and really what you want is some kind of guided serendipity, not total randomness.” The more I use Zite, the less I see randomness, which of course is the point, but as randomness vanishes, so does the magic of discovering something totally new to me.
Next up: music. I wasn’t about to pay for the much-hyped Spotify just yet, but the app (founded by 30-year-old Daniel Ek) will let you use its Radio setting for free on a tablet, which essentially works just like Pandora (P). Enter a song or artist and Spotify plays you tunes it deems similar. I began by typing in Kid Cudi and returned to Zite. While paging through articles, I heard a fair amount of stuff I already knew and liked, plus some new artists I liked, but also more than a few misses. As with Pandora, you may only skip a certain number of tracks. There are ads of course, and many are, well, bad — the majority relentlessly pitched me “behind-the-scenes” footage of Lady Antebellum, a country group I had zero interest in trying. The only obvious, stark difference from Pandora I could identify was Spotify playing me an awful lot of songs by the artist I entered, as opposed to songs by similar artists. This happened with Kid Cudi, Bruce Springsteen, and others. That’s not a negative, assuming you like the artist you select, but it does mean less discovery.
That night, for dinner, Ness gave us Birch and Barley, which my friend had on his list anyway. It didn’t disappoint. We ate with a third friend, who owns a wine shop, and the chef stuffed us with course after course of pasta, fish, and cured meat, plus three different beer samples each time. But in the morning, resolute to have another big lunch despite our food hangover, Ness suggested a Spanish restaurant, Estadio, promising an 81% likeness. The food disappointed, and my chef friend was unimpressed and had to conclude the app had failed us this time.
Of course, a dining app isn’t going to be on-target 100% of the time, nor can a music app avoid playing you some songs that you dislike. Even Zite, which I found most successful at what it does, cannot cover everything for me. It won’t replace the handful of news sites to which I’m loyal and visit every day. On Spotify, even when I hear a great song by a band that is new to me, it often turns out I don’t like the band, and the song I loved was an anomaly. The app will never have as much cred with me as one of my musician friends or authoritative, music-critic peers. The same goes for Ness: Had we ditched it and gone only to restaurants my chef pal already knew about, there wouldn’t have been a single less-than-fantastic meal.
That each app does a decent job, but cannot replace the usefulness of a live person familiar with you and your likes, is no breakthrough epiphany. The more pressing question may be which startup — from these three or from the myriad others already out there — will end up expanding its repertoire to achieve what Reese says is his mission with Ness: “become that trusted source for people to find out the next thing they’ll like.” That sounds like it would encompass a lot more than restaurants.
Reese says Ness could just as easily use its technology to recommend books, movies, travel destinations, or nightlife activities. Then again, Johnson believes the same of Zite. He also posits that coming up with an all-in-one recommendation engine will be inextricably linked to the challenge of perfecting a “Google for the social graph.” Perhaps the final product will be a Google (GOOG) rival that offers what you’re seeking in two forms: one based strictly on your own search history, the other inspired by your personal connections. Take note: It will all rely on big data.
But would we want such a thing? Consolidating the Web is exactly what Facebook (FB) is trying to do; what began as a place for checking people out eventually became your photo dumping grounds and now includes instant messaging, location services, social games, and a commercial marketplace. It remains to be seen whether all of this is annoying enough to cost them users or if people will just give in.
Many avid Internet users prefer to have their various functions in separate silos. I like Kickstarter for funding cool community projects; Twitter for breaking news; Facebook for more personal sharing; and Instagram for photos. I trust each of these entities for the activity it’s perfected. I wouldn’t want a one-stop shop. Moreover, many of these outlets, like multifaceted Kickstarter as well as, say, Tumblr, are appealing precisely because they do not attempt to know you; instead, their users tend to project their own interests onto the platform.
Both Amazon (AMZN) and Netflix (NFLX) (Johnson calls these “the old classics”), which are akin to the “1.0” of predictive technology, still work pretty well and may be hard to beat. Amazon’s recommendation engine relies on a basic formula (despite the highfalutin term they’ve given it, “item-to-item collaborative filtering”) that suggests products to you based on your viewing history, your purchase history, and which related products other customers bought. And it works. The same goes for Netflix, which, as you spend more time choosing movies, becomes quite smart indeed. If these giants of the tell-me-what-to-try space are doing just fine, it may be tough for a Spotify or Ness to simply add more verticals and hit the gas pedal.
For now, I’ll rely on my own human recommendation engine, thanks. My friends and family know me better than any one app ever could. Whether I know myself, well, that’s probably a question for Google, Facebook, and Apple, and their vast piles of data on me — and you.
