“Bots” are pieces of software that rely on artificial intelligence to perform a task. “Chatbots” rely on the influential concept of a conversational UI, in which users interact with bots by initiating a conversation, just as they normally would in a messaging app—”Expedia, I need a cheap flight to London,” and off you go from there, choosing dates, pricing flights, and so on.
Users can interact with chatbots in a number of ways, including voice commands. Accurate, natural-language voice commands spanning multiple languages and accents require vast processing resources that are beyond the budget of most software projects. More commonly in-reach solutions are text-based chatbots—using them is like messaging a buddy on Facebook Messenger, WhatsApp, Slack, or Skype, all of which support their own chatbot platforms and ecosystems.
The Promise—and a Problem
Communicating using natural language with computers, either through voice or typed statements, is an intriguing idea. In theory, enabling users to communicate with software without requiring them to learn an unfamiliar interface, navigate complex layers of screens, or be proficient typists can reduce costs and resources required to connect users with a software service, as well as add elements of accessibility. Saying, “Alexa, turn on the light” or “OK Google, do I need an umbrella tomorrow?” can be comfortable and efficient for many users, even those inexperienced with computers.
In practice, trouble comes when developers try to simulate natural language interactions with chat-based systems. For one thing, natural language input is almost always more verbose than keying or clicking in a command to a menu-driven system, and the time required to enter, parse, and return a natural-language command or response can make such interactive systems seem less than fluid and responsive. Adding to that, it remains an enormous challenge to deliver users a convincing experience, akin to chatting with a human. A user may feel uneasy chatting with a bot that sounds stilted and rote, or with a bot that sounds naturalistic but can’t keep up when a user adopts a more colloquial tone in responses or commands. Natural language processing still has a long way go, and chatting with bots has not yet reached the point of being able to reliably sustain the illusion of truly conversational interaction. Many chatbots try to get around this drawback by offering hybrid chat-menu solutions.
Three items are telling:
- One of Facebook Messenger’s first retail users, Everlane, earlier this month moved away from using the platform and instead has returned to using email to send notifications to their customers. Users can still use the Facebook Messenger chat application to contact customer service, but Everlane’s goals for implementing the bot were evidently not realized.
- Seventy percent of interactions with chatbots on Facebook require human intervention by the chatbot’s sponsor, according to The Information. This has prompted Facebook this month to start to favor hybrids. It unveiled a new menu-driven interface to interact with its Messenger bot that is very much like navigating a “traditional” mobile website. Facebook also issued a suggestion to third-party developers who are working with the platform to: “consider stripping such exchanges down and cutting to the chase by putting the most important features in your menu.
- Interestingly, a Facebook chatbot that is receiving strong positive reviews, Sephora’s Virtual Artist, presents a menu as well as a conversational system. And Sephora has noted that guiding users in how to interact with the chatbot also helps.
Getting Chatbots Right for Your Business
Chatbots have their place in the mobile/Web ecosystem. Getting them right so they enhance your users’ experience is tricky and requires thoughtful consideration of both UX and information architecture, provided by specialists who are strong professional communicators. To learn more, call to set a time to speak with one of our specialists.