At F8, Facebook’s massive annual developer conference, the big news is bots—specifically bots on Messenger, Facebook’s messaging app.
Messenger now boasts 900 million users per month, which presents a tantalizing user base for companies eager to get their wares in front of customers. That’s where the bots come in. Facebook is turning Messenger into an open platform, and any company can now build a chatbot that users can talk with. If you’re an airline, you can build a chatbot to book tickets; if you’re OpenTable you can build a chatbot to take reservations.
“To me it’s about bringing back all the best parts of the interaction between people and businesses,” David Marcus, vice president of messaging products at Facebook, says in an interview. “What we’re trying to build with bots are rich conversational experiences. That’s what we believe will be the future of interactions and services.”
Chatbots seem like a new fad in software design: Microsoft just announced its own suite of bot-building tools; Kik, the messaging platform massively popular among teens, announced one as well. But there are deep reasons why chatbots make sense. The app model has stalled out: People don’t use a ton of apps, and they don’t download many either. The reason is simple: There’s an enormous amount of friction associated with learning about a new app, downloading it, signing up for it, and then remembering you even have it.
The promise of chatbots is that from within Facebook Messenger you can do anything you’d like. For example, you can book an airline ticket for the the first time or call an Uber, with a speed and ease that would be impossible if you were toggling between apps. Moreover, because you’re already in Messenger, there’s no need to sign up all over again if you’re trying out a new service. Messenger already knows who you are, and once you start a conversational thread, your transaction history is right there, threaded into one neat, tidy stream of conversation that you and the bot can access. “We have a two-sided network,” explains Marcus. “There are 15 million businesses using pages and 1.6 billion people using Facebook as their identity. These can now come together in threads that are contextual and canonical. For the lifetime of your interaction everything stays in one place, unlike email.” And, unlike email, your Facebook profile becomes the keystone of your online identity.
The other issue is discovery—exactly the problem that the app ecosystem failed to solve. To that end, Facebook is making sure that the bots are piggybacking onto the existing user flows of the mobile web. If they’re successful, bots will become something like the new Like button: There will be new plugins that companies can add to their websites, which will fire up their chatbots in Messenger. There will also be codes that businesses can slap onto stickers in their real life spaces—much like Kik does—which will allow patrons to fire up chatbots on the spot. Both Facebook Newsfeed and SMS messages will now have a prompt letting you continue a thread inside of messenger.
In that way, Facebook is really hoping to create a new paradigm for the web altogether. You might start off at an airline’s website, but tapping a “chat” button leads you right back into Messenger, where you can complete your reservation. You might start off by browsing your Facebook feed, but end up chatting with CNN’s chatbot. With a tap from almost anywhere, users can be led right back into Messenger, exactly where Facebook wants them. Wherever you are—checking out on a website or reading an article in your needs feed—there will be a new chatbot option that could become a default for millions of people, if they’re designed right.
Designing a chatbot seems perhaps easier than it is.
For Facebook, the process began with a design sprint. Holed up in an Airbnb for three days, the Facebook Messenger team focused on a few problems that they had encountered in the experimental chatbots that they had created for clothing retailers Everlane and Zulily. Perhaps the biggest challenge was teaching people what the bots could even do.
In that sense, Facebook’s increasingly powerful natural-language AI—Wit.ai, which is also being opened up to third-party developers—was a potential hinderance. Let’s say you fired up a bot. What’s the first What does it do? What doesn’t it do? What’s the first think you say? That’s why the experience with bots is less free-form than real chat. You’re presented with a series of options that you click, progressively telling the bot your preferences for how often you’ll be notified or what kinds of information you’d like to receive—”Like a Goosebumps choose your own adventure book,” explains Jeremy Goldberg, a product designer for business and platform at Facebook.
But there are some particulars about how the chatbot itself should behave. Facebook boils these down to four design guidelines, that all bots should adhere to: That the bots should be conversational, issuing messages in a natural tone of voice and short, pithy responses paced quickly but not so quickly that the bot is oppressively present–like a chat partner hanging on your every word and expecting just as quick of a response. The responses themselves can be interactive bubbles. Some let you horizontally scroll through a list of sweatshirts via Everlane or a list of popular stories via CNN. Others present options that you can tap—for example, you could share your present location with a weather app like Poncho—thus quickly making your your conversations more useful.
Chatbots in Kik and elsewhere work much like this already, with guided options that push your chat along. But Facebook believes that the future will yield deeper and deeper integrations between the chatbots and the sensors and data on your phone. “We will keep pushing that forward. More and more kinds of messages and information that you can exchange,” says Austin Bales, product-design manager at Facebook. Moreover, the natural language AI will get better, as bots hoover up more and more data about the things people say and the questions they ask. Over time, according to Marcus, this mix of natural language chat and tappable bubbles will yield a chatbot UX that feels increasingly fluid and intuitive.
Facebook, for its part, believes that chatbots are already surprisingly intuitive to use. “It’s like before you order a cab or takeout on your phone. Before you did it for the first time, you probably thought that calling a phone number or using a website was just fine. But then you use an app and it just makes so much sense,” says Bales. To which Goldberg, his colleague, adds: “There was an ‘aha’ when we showed people the prototypes. They hadn’t thought about interacting with businesses the same as you would with a person. But then it just seems like, ‘Of course you should be able to do this.'”
Which makes sense. Then again, we’ll see how well users cotton to chatbots, which have, after all, been around since the days of IRC. Only users will decide if they really are faster and more fluid than apps; the proof will be in both the numbers and the nuances of how useful the chatbots prove to be in aggregate. Facebook probably has to get it right, if for no other reason that as a hedge. With users reportedly sharing less and less on Facebook every year, it makes sense for them to create a business-friendly platform where users already are: Happily chatting away, devoting as much time as ever to talking directly with friends on their messaging apps, with less and less need for Facebook proper.