6 best practices for IT pros when building a chatbot
There's a lot to consider when building a chatbot. Beerud Sheth, co-founder and CEO of Gupshup, broke down six best practices for IT pros at the Global AI Conference in Boston.
BOSTON -- If businesses are to "go where their customers are," that can only mean one thing: They need to embrace -- and create -- meaningful conversational experiences, according to Beerud Sheth, co-founder and CEO of Gupshup, a San Francisco-based platform that helps enterprises build chatbots and conversational experiences.
In order to truly reap the benefits of conversational experiences, businesses must master AI-based chatbots, Sheth argued at this week's Global Artificial Intelligence Conference. According to Sheth, chatbots are easier and more natural to use than apps -- they're also inherently easier to personalize. Bots have already begun transforming enterprise support, sales and marketing functions across industries.
But where should the IT pros begin? Sheth broke down some beginner best practices and considerations for designing and building a chatbot.
Define the scope of the chatbot
The place to start is with the definition of chatbot, which varies depending on whom you ask. Sheth said his definition of chatbot encompasses both text- and voice-based conversational systems.
The first big challenge IT teams face when building a chatbot is keeping their ambition in check and their goals modest.
"The moment people hear AI, their minds start running, and they get too ambitious," he said. "But we're still quite a ways away from general-purpose AI that can handle every user query."
The key is to focus on doing a few things and doing those things well, he said. IT pros can add more capabilities incrementally, but they should always stay within the defined scope of the bot.
Set the right expectation with users
The technical need to limit the scope of the chatbot also means it's necessary to accurately set user expectations. Revealing limitations upfront goes a long way in enabling user satisfaction, especially when building a chatbot for business, Sheth advised. "Overpromising and underdelivering" tends to be a leading cause of user disappointment or frustration, he said.
He referred to this as the "Siri problem." He explained that when Apple's AI assistant was first introduced, it encouraged users to "ask it anything." But, of course, there were a lot of questions it couldn't answer. Setting the right expectations would likely have helped users ask Siri more relevant things and better highlighted Siri's strengths.
Build guide rails
Even after properly defining the scope and introducing the technology's scope correctly to users, people will still ask questions the chatbot won't understand. In these situations, it always helps to have the chatbot not just say, "I don't understand," but also offer suggestions that can help get the conversation back on track.
"If the chatbot just leaves it at, 'I don't understand,' the user is left hanging and doesn't know what to do," Sheth said. "But having it say, 'I didn't understand, but here's what I can help you with -- A, B or C,' just makes it a lot easier and prevents the user from going completely off course. You can keep guiding the conversation back to where it makes sense."
Set a consistent tone and personality
Add in small, delightful moments that can lighten the experience and disarm the user, so even if there's a problem, they're more forgiving and tolerant.
Beerud Shethco-founder and CEO of Gupshup
Each of our friends and family members has a unique personality.While chatbots are far from family, Sheth has found that humans still expect a certain -- and consistent -- level of humanness in chatbots. When building a chatbot, he advised IT teams to explicitly think about and deliver one that has a consistent tone, whether that be professional, casual, funny or chatty. Integrating colloquialisms that fit the tone come into play here, he added.
"Add in small, delightful moments that can lighten the experience and disarm the user, so even if there's a problem, they're more forgiving and tolerant," Sheth said. "Just a tiny bit of humor goes a long way -- especially when the bot reaches its limit."
Structured vs. unstructured
IT pros also have to decide whether they want structured or unstructured conversations in the chatbot. Structured conversations are highly scripted and in a dynamic FAQ-like format, restricting the kinds of things that the users can do, including inputting freestyle queries. The advantage is the business can control the whole conversation. The disadvantage, of course, is you may not have anticipated all the things the users might want to do with the chatbot.
In unstructured conversations, users can type in anything, in any way they like. This allows IT to broaden the scope a little bit, but it's harder to implement, and the risk of user frustration is a little higher, Sheth said. He actually recommended companies have both features available to users in some capacity.
Voice vs. text
Should a chatbot be text-based or voice-based? Sheth said IT pros shouldn't worry about that when designing a chatbot; the key is to build a bot that can be integrated on multiple channels.
"Build a bot with a consistent core logic, and then you can selectively publish on particular channels," Sheth said. "That's critical for maintainability. As you build future versions of the chatbot, if you build it uniquely to each channel, it can be quite challenging to upgrade and maintain as you add new features and so on."
