Chatbots are supposed to help businesses stay in touch with customers. But can they help people stay in touch with friends too? I built one to find out.
For the past few days, I’ve been working with Smooch.io, a startup that helps businesses launch messaging services without needing a lot of technical knowledge. My coding experience is limited to the HTML I remember from building a GeoCities site in middle school … so Smooch.io seemed ideal.
The Canadian startup provided me with code for a basic bot on Github and an easy way to plug that into various chat platforms, such as Facebook Messenger, Line, Zendesk and Slack.
I chose Facebook Messenger because that’s what most of my friends use to chat. I used Heroku (at Smooch.io’s suggestion) to run my code. Everything was free.
After working out a few kinks, I successfully deployed a chatbot on my Facebook page in about two hours. But I ran into a problem as soon as I got my first message: The bot didn’t really serve a specific purpose.
When Facebook announced the launch of its Messenger bot platform in April, CEO Mark Zuckerberg said bots would remove the need for people to call customer service.
Retailers like 1-800-Flowers use a Facebook chatbot to help people place orders. Hyatt lets travelers book hotel rooms. I just wanted a bot that could talk to my friends, but the bot I built couldn’t say much.
The plug-and-play software came pre-programmed with a few keyword responses, such as “Awh, shucks! I love you too!” if someone messaged me “I love you.”
I added more replies so the bot would sound like me and hold a basic conversation about the stories I enjoy covering. For example, if someone told the bot they were my friend, it would trigger this response: “Hello, friend! What’s goin on.”
Despite creating over two dozen more responses, I couldn’t customize my bot to run smoothly.
My friends often got stuck with the same reply: “I’m a pretty basic bot right now. Try asking me how I’m doing or what I’m working on. Are you a FRIEND or PR PERSON?”
Keith Brisson, CEO and cofounder of Init.ai, says conversations are hard to program because computers still have a difficult time processing natural language.
“Computers aren’t going to be like people anytime soon,” said Brisson.
He says his company can help businesses build “conversational apps” by using AI, similar to what IBM and Microsoft are trying to do. But I couldn’t test Init.ai because its product wasn’t ready yet. And even though Facebook says AI is a big part of what makes its bot platform powerful, I clearly didn’t implement my bot to take advantage of the technology.
My lack of coding skills and unclear focus meant that my bot was flawed from the start — especially from Facebook’s perspective.
The company advises developers to design their bots with specific goals in mind: What actions should people take when using the bot? What are all the paths people can navigate during their interactions?
“Focus on doing a couple things really well,” the guidelines read. “Doing too much creates confusion and dilutes your experience.”
Facebook also told me that, at the moment, people tend to find bots more useful if they act more like apps (with clear buttons to tap) and less like conversational, free-form text services.
Take Poncho, a weather bot, for example. Poncho sends weather forecasts as a message in Messenger and gives people set options to choose from when requesting more information.
“Ask me about the weather anytime. If you’d like, I can also send you forecasts twice a day,” Poncho says. Then it follows up with two options: “Set Notifications” and “Nah, later…”
Businesses that have complex customer service needs, along the lines of what I wanted to do, may want to wait until bots can go beyond performing simple routine tasks.
“We launched the Messenger bot platform less than two months ago… [It’s] still in beta,” a Facebook spokeswoman said. “We’re looking forward to continuing to get feedback and to making improvements.”
