Facebook Messenger Bots, while still in early experimental stages, offer a window into the future of mobile experiences.
Where Are We Today?
If you think Facebook had a brilliant idea with Messenger Bots, you would in part be right. Facebook was not responsible for ideation, though. Globally, Facebook is late to the game. Telegram, an encrypted Messaging app offering bot integration, is now well over 100M active monthly users, adding 350K new users per day . Kik, another Messaging app offering bot integration claims 300M registered users . Kik is directly taking on Facebook in the “Messaging Wars” .
While Facebook says the Messenger Bots platform is just an “experiment ”, their strategy is evident. If it’s possible in the US market to centralize activity from a bunch of different apps and businesses inside Facebook Messenger, they will do everything in their power to make that happen.
Venture Capital firm Andreessen Horowitz posted this article on ‘what happens when one app rules them all.’ Author Connie Chan discusses how WeChat morphed from a Messaging app into a collection of millions(!) of lightweight mini-apps which all live inside WeChat. This is exactly where Facebook wants to go. Facebook Messenger bots are effectively mini-apps.
Facebook’s approach is however quite different from WeChat. WeChat offers an app-store for mini-apps inside WeChat. Like on iOS or Android in the US, all the apps have different user experiences, different interfaces; they all look and function the way the creator of the app wanted them to.
Facebook Messenger, on the other hand, has created a user interface library which puts many constraints around what developers can and cannot do with user experience and user interface design. Facebook’s imposed controls are not only designed to limit what is possible in their platform (thus keeping their “experiment” more contained, focused), but also to ensure that users always have a consistent look and feel and good experience.
How Are Things Going with Facebook’s Messenger Bot Platform and the World of Bots?
Facebook is just getting started. Its wit.ai platform and Messenger Bots platform provide developers a method through which integrated experiences can be created and delivered to consumers inside its 1-billion+ user Messenger app.
In typical Facebook fashion, to drive early adoption with developer partners, Facebook has made their APIs simple and easy to use. My engineering team at Stuzo had no trouble building functional prototypes using Facebook Messenger Bots and wit.ai with minimal ramp-up time.
Several forward-leaning companies have jumped in with Facebook to play their role in this grand Conversational Experience “experiment”.
How Are Things Going with Companies that have Created Bots on Facebook’s Platform?
To start, not great. I’ve been evaluating bots released to Facebook Messenger from notable companies, government, and startups. Here’s what I found:
The Promise: Most would agree that booking airfare or a hotel is not an easy or pleasant experience. Having a virtual assistant to make the process easier — more like having a conversation than using a complex user interface — is a promising proposition.
The Reality: The bot doesn’t know how to make sense of basic language yet. This isn’t a failure of AI. In fact, AI is already quite good at understanding basic language. This is a failure on Expedia’s part to understand how people think and talk.
Examples: I started a search for airfare in the Facebook Messenger Expedia bot. After pausing my activity, the Expedia bot came back with a message asking me to restart my search. When it did, I responded with “Stop” because I did not want Expedia to send me messages at random times without me asking. What happened when I said, “Stop”? Expedia said, “OK, let’s start a new search”. I responded with, “No thank you.” Expedia answered with, “OK, Thank You BLVD, Darlington, SC.” The bot clearly did not know how to handle my simple instructions. I did, however, learn of a street named Thank You in a place called Darlington, SC. Which is something. Screen shot below.
What Needs to Be Better: The bot needs to know how to handle basic commands. “Stop” should mean something to a user’s experience. The bot should also understand when a user is getting frustrated (a la my experience of giving the bot two negative responses in a row — “Stop”, then “No”).
U.S. President Barack Obama
The Promise: President Obama is arguably one of the wittiest presidents to ever grace the oval office. Presumably, one could have a lot of fun conversing with an Obamabot. Or more seriously, asking the Obamabot about U.S. foreign policy and perspective on involvement in Syria, or asking about the latest headlines from the White House, or asking about any upcoming scheduled public appearances, and so on.
The Reality: I can’t talk to the White House or President Obama at all. In fact, I haven’t been able to get the White House bot to respond to anything; even a simple “help” command.
Examples: When asking the White House bot, “What is your stance on Syria?”, the response was to click the “Let’s Go” button above. I then asked for “Help.” Same response, “Click Let’s Go.” OK, fine. I clicked “Let’s Go.”
Next, the bot asked me what I would like to say to President Obama. I asked, “What is your stance on Syria?” The bot asked me if that was my whole message. Again, frustrated that the bot cannot handle a simple question or at least acknowledge that there was a question mark at the end of my sentence, I clicked “No”. It responded, “No problem! What else would you like to add?” I said, “Change my statement.”
The bot clearly did not understand and again asked me if that was my whole message. Now I’m thoroughly confused. Is the bot going to send a message of “What is your stance on Syria. Change my statement,” to the White House?
I wanted this terrible experience to end. I type in “Stop.” Again, failure. The bot does not know how to stop working. If the U.S. Government knows how to do anything right, it is to stop working. (Sorry for injecting politics into my article. I will further abstain.) Screen shots below.
What Needs to Be Better: VentureBeat did me a solid on this one and already wrote an article on How to Build a Better Bot than President Obama’s in 15 Minutes.
The Promise: I bought a new pair of eyeglasses from Blenders. My Blenders account/profile was already integrated with Facebook (that is, I use Facebook to log in to Blenders). Upon successful checkout, I got a notification on my phone from the Blenders bot inside Facebook Messenger. Interesting. Turns out, my order confirmation was available inside Facebook Messenger. My initial reaction: This is useful. I like having it here. Having Facebook Messenger integrated with places I normally do commerce online could make for a nicely integrated experience.
The Reality: When I received the confirmation message in Messenger, I wasn’t able to tap/click on the visual for my order confirmation. At first I thought it was a bug. ‘Of course I should be able to tap on the order confirmation to view my order details,’ I thought. This was not a bug. It was by design. Poor design.
Examples: When my order shipped, I also received a note in Messenger that had my shipment tracking number and links to view my order and track my shipment. Again, I liked this. Another good opportunity to test the bot’s ability to process simple language. I said, “Thanks”. Apparently, Blenders did not want to talk to me anymore (sad face). Screen shots below.
What Needs to Be Better: Something that looks like a button or an interactive element in a mobile app or on a Website, should also be an action in the bot. Also, there is no reason why an automated bot should ever be offline. Even if the Webservices powering the bot are offline, a simple “we’re down for maintenance” message would have been a much better user experience than receiving a default red Facebook error message.
It’s Not the AI; It’s the UX
From my perspective, bots can do a lot better. I have yet to see a bot experience that excites my nerdy, tech sensibilities.
Given the suboptimal bot experiences I’ve had to date, one might assume that the problem with these bots is the maturity of artificial intelligence. AI maturity is not the problem. It’s only part of the problem. And this partial problem with AI, is that we humans haven’t yet figured out how to harness its true power and potential. That’s only because we haven’t taught AI enough yet. Once we have, AI will be able to think at an alarmingly grand scale.
The problem with suboptimal experiences we have today, lies in how we approach building complex conversational logic. User-driven or empathetic design practices are nothing new. Stanford covers empathy-driven design in its Design Thinking school. Great design has been around for millennia. However …
Conversational UI Has Not Been Around for Millenia
To design a beautiful conversational experience, one needs to start with the user and consider: What does the user already know how to do? What does the user want to do? The answers are simple: The user already knows how to talk. The user wants to get simple tasks done faster and more intuitively.
While these answers are simple, the design process and logic required to create the conversational experience are anything but. Consider this. A conversational experience with a Facebook bot currently allows for three types of user actions, listed below. 1 and 2 are well known. 3 is groundbreaking:
- Click/tap on presented option
- Use a navigational menu
- Talk with natural language text
A Forward Look into Conversational User Experience
When it comes to user experience, for 1 and 2 (above), we’ve been designing this way for 20+ years now:
The 3rd user action, talking to a bot with natural language text, is totally new. As you can imagine, this gets complex fast. Imagine a Facebook Messenger bot starts with a simple question, such as “How are you feeling today?” The bot presents three options to the user, as shown below:
Based on how the user responds, “Great, Just OK, or Bad”, the bot can present the user with any one of the three user actions in response: 1. Option to click, 2. Menu, 3. Natural language. Let’s see what happens when natural language is inserted for the option of having the user describe to the bot why they feel bad:
Next, an AI engine and a lot of logic is needed to handle the user’s response, as shown below:
Because of this complexity, what we need is a new way to approach designing for conversational logic.
The first step is to define a new conversational lexicon; that is, a new set of terms and definitions that describe how a user interfaces with an artificially intelligent messenger and a new way to systematically organize complex logic trees (like shown above). The lexicon must define how AI experiences should respond with simple text, with text-based feedback loops (Q&A cycles), or with rich-media feedback loops containing text, images, calls to action, video, audio, and interactive elements.
On this front, Stuzo is advancing quickly. We’re already building bots for Fortune 100 companies and in order to do so, have created our own proprietary AI UX lexicon. Armed with a common language and common design practice for how to approach conversational experiences, we can build a great experience around existing AI that’s already equipped to handle complex queries.
If your organization is ready to unlock the potential of Conversational Experiences, contact us. We’d love to hear about your challenge.
 According to Kik’s Press Kit: https://www.kik.com/press/
 According to Forbes: http://www.forbes.com/sites/parmyolson/2016/02/10/kik-bots-messaging-facebook-wechat/#21ebe7242571
 I know, as I’ve talked to Facebook executives directly about their Messenger bots strategy.