voyager624 - Fotolia
Published: 05 Jun 2018
Chatbots seem to have all the answers, but they don't get smart all on their own.
Designing chatbots that are informative and persuasive means having a clear idea of what your virtual agent will convey to customers about your organization. Getting to that point takes time, patience and an understanding of the nuances of language. It also requires a close working relationship with a software developer.
"It's a new medium. The answers are not obvious and it's not instant," said Beerud Sheth, CEO and co-founder of chatbot platform Gupshup. "You have to do some trial and error."
No organization has yet rolled out a perfect chatbot. Like humans, chatbots misinterpret typed and spoken words -- and make mistakes. Being prepared for those inevitable setbacks is one of the first things your organization should realize when working with a virtual assistant developer.
"We don't go for the moon. We don't encourage that," said Andy Peart, chief marketing and strategy officer for the natural language platform Artificial Solutions. Still, the chatbot must have an acceptable level of functionality so it doesn't chase away customers, he added.
Treat it like it's human
When designing a chatbot, don't expect it to provide much value if you want to use it for the sake of showing off. Have a firm idea of how it will be used and why. Plan for its role as you would actual human employees. Most Gupshup clients, for example, want their prospective chatbot to help customer growth, followed by sales enhancement and then customer support, Sheth said.
Once you map out the chatbot's duties, you'll need to select a team that will work with the developer from design to implementation to managing and monitoring the chatbot's everyday use. "We do not need technical or linguistic experts on that team. Rather, they should be people with a deep knowledge of the business," said Mariana Martinez, former content and marketing specialist for virtual assistant platform Aivo.
A business-minded team will be best-suited for answering critical questions that will align the chatbot to its audience, Martinez explained. Those questions should include: Who is your buyer? Why do they buy? What do they buy?
Brand a personality
Most companies already have a defined image, color, logo and sometimes an avatar in mind when designing chatbots, Martinez said, adding that design, tone, language and vocabulary can shape a persona. While the chatbot's personality should match the company's, virtual assistants also can be designed to fit the personality of customers. Gupshup, for example, created a chatbot for Wrigley's Skittles candy that mimicked the colorful and fun marketing messaging of Skittles. "It takes a lot of brainstorming," Sheth said. "It's critical that the personality comes through. The brand owner has to be involved in the process."
For the website All About Jazz, the persona of its Facebook Messenger chatbot tries to align with the wide perspectives of the readership. Newcomers will type basic questions about jazz -- inquiries about an artist or a concert date -- while discerning longtime fans might forward queries about an album's liner notes. The trick is to have the chatbot's personality reveal itself in language, with words and phrasings that don't turn off users.
The personality of the chatbot can also be found in its name. All About Jazz founder and publisher Michael Ricci chose "Bix" in homage to 1920s jazz cornetist Bix Beiderbecke. Old-school jazz fans will get the connection, while new ones will think it's simply a cool name, Ricci said.
Talk it up
The words a chatbot uses to communicate and the words it will be "trained" to understand are methodically worked out between client and designer before they're implemented into the software. This might be the most difficult and time-consuming phase of development. After all, a chatbot is supposed to "chat," whether it's verbally through a voice-activation platform or through typed messages via text or social media channels.
Chatbot conversation is governed in one of two ways: structured and unstructured. Structured conversation has the chatbot putting a gate around a chat by using selected words so that the human, stuck with limited choices, won't drift far. This kind of chatbot can display responses with encoded links to lead web and mobile visitors directly to what they (should) want.
"When you do a structured conversation, there is no chance for misunderstanding," Gupshup's Sheth said. A structured conversation, he added, can be crafted in more than 200 languages "very easily."
If your company wants to produce a chattier chatbot, it will have to work without language parameters. The chatbot's responses will come from a repository of phrasings that are assigned to the words used by customers. It takes years of conversations for humans to almost fully understand one another, and that's not always guaranteed because of a misplaced word or an unusual idiom. Chatbots are like toddlers and just starting to "understand" the structure of sentences.
"It gets a lot harder with unstructured," Sheth said. He recommends carefully choosing all the words the chatbot will use through a structured language format and gradually giving it more to handle.
Artificial Solutions' Peart similarly warned clients not to get their expectations too high in this early age of chatbots. "The problem with chatbots is they're pretty dumb," he said. "They follow a linear path and can't stray."
When designing chatbots, don't leave them hanging in the wind when they fail to understand human voice commands. Using natural language development and analytics, chatbots will become attuned to words and phrases outside their usual repertoire because they're gaining experience in conversations.
"Customers might not be happy with answers even though it's working well," Peart said. "Companies can mine that data. They can see when people are getting annoyed and what caused it. You can fine-tune that."
Chatbots will continue to "learn" from their mistakes, because they don't appear to be going away anytime soon. With an opportunity to engage customers on a basic level and cut the costs of human representatives, chatbots will become an even bigger part of CRM.
Gartner estimated that in two years, 25% of customer service and support operations will integrate virtual customer assistant or chatbot technology across their engagement channels. "We've done surveys looking at voice assistances," he reported. "It's growing. People increasingly expect to speak to devices in human ways, and soon, the expectation will be that every enterprise has one."