“Do you want to get dinner next week?” In my experience, this question inevitably leads to a long quasi-negotiation, conducted via email or text message. It usually goes something like this:
Friend 1: Sure! Do you have a place in mind?
Friend 2: Not really! I’m up for anything. Maybe we could meet in Brooklyn after work?
(At this point we’ve narrowed it down to a borough of 2.6 million people. Doing great!)
Friend 1: Sounds good. Are you in the mood for going someplace nice or something more low-key?
Friend 2: Hmm, I don’t know. Maybe we could go with pizza to split the difference?
Friend 1: Oooh, let’s do pizza. There’s Franny’s, Saraghina, Motorino, Roberta’s …
Friend 2: Well, Roberta’s has that outdoor bar area you can wait in. If the weather’s nice, we can go there, and otherwise, we can do Franny’s?
Because of two people’s desire not to be perceived as pushy and/or their general indecisiveness, a decision that could have been made in two steps ends up taking six. And if there’s a Friend 3 involved, expect to add at least another couple of steps—unless Friend 3 is vegan or deathly allergic to gluten, which usually helps narrow down the choices. (By the way, the above dialogue is based on an actual recent discussion. We went to Roberta’s.)
Deciding where to eat, drink, relax, and chat with friends should be a pleasure, but instead it’s an engine of hesitancy and chagrin. As a result of that hesitancy and chagrin, you often end up going to the same handful of tried and true restaurants instead of branching out. What if technology could solve this problem by collecting a party’s various dietary, monetary, and atmospheric preferences and producing a restaurant that will delight everyone?
Ness was an app that promised to do just that. (I say “was” because it was acquired by OpenTable in February and subsequently shut down operations in order to incorporate its algorithm into OpenTable’s framework. More on that later!) Ness looked like the Netflix of restaurants: It invited you to rate restaurants you’d been to, then suggested other restaurants based on ratings made by people with similar preferences to yours. Taking a cue from OkCupid, Ness expressed its prediction of how much you’d like a restaurant as a “like percentage”—the higher the percentage, the stronger the recommendation. Ness’s “Recommendations With Friends” feature also figured out where your preferences and those of your friends overlapped.
I decided to put Ness’ algorithm to the test by recruiting colleagues to try a restaurant for the first time using the app. I made them promise that they would go to whatever place Ness recommended for us, and 14 brave co-workers signed up.
Which is where I ran into my first problem: Ness’ group recommendation algorithm accommodated 10 people maximum. I split us randomly into two groups and then immediately hit my second problem: Ness could recommend a restaurant for each group, but it couldn’t help us figure out what night we were all available to dine. We ended up using Doodle to find a night that worked for every member of each group.
Now for the moment of truth: I plugged in the names of the eight people in my group, chose “dinner” as our meal, and limited the geographical area to the West Village, near Slate’s office. (I also limited our price range to $ or $$—Ness’ ratings go up to four dollar signs. We work in journalism, not finance.)
With all this information, I expected Ness to spit out the name of the single restaurant with the highest average “like percentage” for all eight of us. Instead, it gave me a list of restaurants; confusingly, the list changed every time I refreshed the page. The main thing that had attracted me to Ness—that it would eliminate all decision-making—turned out to be a vicious lie. I still had to make a decision!
I chose Aria Wine Bar, because it showed up high on the list every time I refreshed it, and because it seemed to have a pretty high “like percentage” for most of the eight of us. But it turned out my boss had already been there—Ness couldn’t sort for novelty. I did find a pizza place on the list that no one had tried, but then an editor in my group let me know that she didn’t eat wheat or dairy. Ness didn’t allow people to input their dietary restrictions, either, even though dietary preferences tend to be the major determining factor when it comes to group dining.
Long story short: After three people bowed out at the last minute, five members of our group went to Aria Wine Bar. (Yes, I did a new search for recommendations for just the five of us. Ness still really wanted us to go to Aria Wine Bar.) It was hard to communicate with one another there, both because of the high volume of the dance music playing over the loudspeakers and because we were seated mere inches away from our nearest neighbors. We ordered appetizers to share and ended up with an unconscionably cold slab of mozzarella; our pasta entrées were bland at best. Afterwards, we all agreed that we would not recommend Aria Wine Bar to a friend.
So Ness did a great job of recommending a restaurant none of us would like. Instead of tailoring its recommendation to our preferences, it seemed to target the lowest common denominator among the five of us.
Was our experience a fluke? It was not. The other group of Slate staffers who tested Ness on a different night also got Aria Wine Bar as a top recommendation. There are dozens of good, reasonably priced restaurants in the West Village, but if you wanted to plan a dinner in the West Village for five or more people chosen randomly from the census, Ness would tell you to go to Aria Wine Bar. Luckily for my colleagues, there was a long wait, so they ditched and went to neighborhood institution the White Horse Tavern instead. My colleague Bryan Lowder reports that they had a lovely time.
“We all agreed that the funny thing was that Ness would never have suggested this place to us based on our ratings,” Bryan told me afterward. “I feel like very often the best dining experiences happen by chance like this—Ness can’t account for that.”
A technology that does for restaurants what Netflix does for movies might be a pipe dream. Compared to restaurants, movies are discrete experiences; they’re exactly the same every time you watch them. You might choose a different movie depending on your mood, but there are only so many criteria you can use to evaluate a film, and Netflix has done a good job of breaking them all down. (Hence ridiculously specific categories like “critically-acclaimed emotional underdog movies.”)
Restaurants are harder to put into boxes. Your experience at a restaurant will vary depending on what time you go, where you’re seated, whom you’re with, who your server is, how crowded it is, and what you order. Restaurants can get better or worse over time. Menus change. Sometimes you can put your finger on what bothers you about a restaurant. (“The mozzarella was unconscionably cold.”) Sometimes you can’t.
That said, it’s very, very easy to imagine an app that does a better job of quantifying preferences than Ness did. That’s why it came as such a surprise to learn that OpenTable, the online restaurant reservation company, shelled out $17.3 million for Ness’ proprietary algorithm. So far, OpenTable and Ness have been vague about how they’ll combine their technologies: Presumably, OpenTable will add Ness’ algorithm to its reservation database in the hopes of being able to recommend new restaurants to users and let them make reservations online. In this way, OpenTable would compete more directly with Yelp, which just launched its own online reservation system.
There’s time for OpenTable to iron out some of Ness’ kinks and to add some potentially helpful new features. As for me: I’ve gone back to the old-fashioned, six-email-negotiation technique. This technique is awkward and time-consuming, sure—but after trying Ness, I’ve come to see it as a necessary evil.