What’s the best way to experience a new city? Most people would say with a local as your tour guide with similar tastes as you. Plenty of startups have tried to duplicate this personalized discovery and recommendation experience, including Foursquare, Nara, and dozens of others. But none have quite nailed it yet. Be it the user experience of getting to those recommendations, or the recommendations themselves, local discovery remains largely an unsolved problem.
Russian startup IO thinks it has a novel solution in its linguistic recommendations engine which allows users to talk, via type, to the app as if it were another human being and receive tailored recommendations based on the situation. For example, dinner recommendations for a romantic evening should be different than those on a work outing.
The company is in the process of relocating its operation from Moscow to New York, including transporting its 10 person engineering team, and will launch a US version of its iOS app – focusing initially on New York City – in beta in September, according to co-founder and CEO Zhenya Kuyda. A full commercial launch is currently slated for November.
“New York City is the ultimate market for restaurant recommendations,” Kuyda says. “If we can make it here, we can make it anywhere.”
The novel aspect of IO is twofold, according to Kuyda. The first is the app’s natural language interface. IO accepts and seamlessly parses queries, such as, “I’m going to dinner with my girlfriend, and I’m looking for something romantic but not too expensive.” The app might react to such a query with an equally conversational response, like, “Ok, are you in the mood for Italian or Sushi?” or “How do you feel about hipsters?” Next, it might ask, “Would you like to make a reservation at 8:00 or 8:30?” (If it’s not already obvious, the company is initially focusing on restaurant and bar discovery.)
The second unique aspect is the way IO learns about a city’s dining options and individual user preferences. When launching a new market, the company first ingests restaurant descriptions and public user reviews on sites like Yelp, Foursquare, and food blogs to ascertain information like ambiance, menu, ratings, and price. With this lay of the land established, the app then continues to learn about individual users with every interaction. For example, a user who says “I don’t like Italian” or “my girlfriend is allergic to shellfish” will see their future recommendations adjusted to reflect this information.
“We’re not a personal assistant or a Siri competitor,” Kuyda says. “Siri is not there for conversation, it’s there to answer one question. It will answer the same question the same way ten times in a row. But IO will answer the same question ten different ways. This looks like a more human conversation where the dynamic goes back and forth and sometimes the best answer to a question is another question.”
Therein lies the beauty and the challenge of IO. The app has been in a private beta in Russia for much of this year and has been getting positive reviews, according to Kuyda. But the company’s English language version is currently in early alpha stages and will take some time to collect enough data to perform up to par. Early users are likely to find that the app is still learning the city and nuances of the language, which may slow the app’s adoption among a notoriously difficult New York dining crowd.
But even once IO knows New York like the back of its hand, it still must learn each individual user. This means that the first impression of the app, while “cool,” will always offer just a glimmer of what the platform will eventually become after committing a bit of “getting to know you time.” In that respect, IO will be a lot like dating. The company will hope that users see enough of an initial spark to invest the time to move beyond first impressions.
This is a problem not uncommon to Foursquare, Nara, and IO’s other competitors. Each has solved the first impression problem in its own way. For Foursquare, the solution was years of gamified check-in data to help map a city and learn individual users’ preferences, combined with ratings, and a new user onboarding experience that involves selecting from a list of preferred food and entertainment genres. For Nara, the magic lies in a user inputting data around favorite spots in their hometown so that the app can recommend similar spots in other cities.
If IO wants to compete, the company will need to prove it has its own clever way of shortcutting this onboarding and getting to know you process. To its credit, the app gives new users a bit of a guide, suggesting possible follow up questions mid-dialogue, such as “How’s their menu?,” and “Is it loud?” But this is just a start and the company will likely need to go further in lowering barriers to entry.
“People don’t want to be filling out forms when when they start to use a new app and they don’t just want the same restaurant for all occasions,” Kuyda says. “We see ourselves as the interface of the future. We can eventually add in all the services and content providers you can think of. This is an interface that can be common across all services. We plan to open other domains in the future, such as hotels, travel, and leisure. We will roll out slowly to see how people react, and eventually add more features and topics. We’re focused on building a great solution to the human machine interface problem.”
From a feature standpoint, IO will continue to evolve as well, Kuyda says. The company is in the process of adding booking and delivery options to its US version, features that already exist through partnerships in Russia and have proven popular in that market. The company is also working on a multi-user chat experience that will feature recommendations based on blended user preferences, which is something I have yet to see in any competing products.
IO raised $2 million* from primarily European VCs earlier this year, Kuyda says, while declining to name names because “the round is still open and may grow a bit larger.” The company was bootstrapped for its first several years of R&D in Russia and established a distribution partnership with RAMBLER&Co, one of the country’s largest Internet conglomerates. In the US, IO plans to go it alone, with the exception of partnerships around features like booking and delivery.
[*Update: In an interview preceding this article IO CEO Zhenya Kuyda said to Pando that the company had raised $2 million, in pursuit of a $2.5 million round. After publishing, a company spokesperson clarified that it in fact has only received commitments for that $2 million amount, but has yet to close on any part of this funding. Additionally, the company will complete its move to New York only after closing its funding round, which it anticipates doing in September.]
With its launch around the corner, it’s not surprising that IO is focused on user experience and driving adoption rather than monetization. But the path to revenue is fairly obvious. The company could eventually choose to offer native and display advertising, and has the potential to earn lead-gen commissions from its booking and delivery partners. Those, however, will be decisions for a later date.
One upside of the new market expansion, according to Kuyda, is that the linguistic engine actually works better in English than it did in Russian, due simply to the complexities and structures of the two languages.
With its natural language interface and machine learning algorithms, IO is somewhat reminiscent of the recent hit movie Her, which explores a dystopian future in which a man’s best and closest relationship is with a machine assistant. Sure this is just an early glimpse at what such a future might look like, but it’s also a reminder that it’s not necessarily that far off.
“In just our first two months in beta in Russia, the average user sent 200 messages to the app over that time,” Kuyda says. “People write things that surprised us, like, ‘thank you’ or ‘you’re a bitch’ or ‘talk to me please’ – although that was more so, girls who seem to be more emotional. Initially, we expected people to use it more like they use Google and just type ‘pizza,’ but that hasn’t been the case. But, if you want to get a fully personalized app that knows everything about you, you need to talk to it. This is just the start of that conversation.”
[Illustration by Brad Jonas for Pando]