Voice User Interfaces and How To Write For Them, Part II
Mar 21, 2017
In my last post, I announced that Centerline is building a voice-only chatbot. When finished, it will join our recruitment team at hiring events in order to help them answer “softball” questions from interested candidates.
For job seekers, we want to deliver an experience that’s intuitive and seamless. We’ll have to overcome some common challenges associated with VUIs. In my next few posts, I’ll share the best practices we’re applying to overcome those challenges.
To start, I’ll tackle how to help users discover the features of a voice-only interface. How do we make the VUI intuitive? How do we help users complete a task or start a conversation?
Command-and-Control vs. Conversational Interfaces
First, it’s helpful to characterize the interaction that’s most likely to occur: command-and-control, conversational, or both? In a command-and-control situation, the user interacts with an agent to complete a task. This is the type of relationship most users have with digital “personal assistants” like Siri or Google Assistant:
- “Text mom: ‘please stop liking all my Facebook photos.’”
- “What’s the temperature in Longyearbyen, Norway?”
- “Set a reminder for 8:30 a.m. labeled ‘put pants on.’”
Typically, command-and-control interactions are brief (one or two turns in the conversation).
Contrastingly, the conversational style is just that: a conversation. Our chatbot will conform to this conversational style. For instance, a candidate might ask:
- Candidate: “What is a typical workday like?”
- Bot: “With the exception of ending them at the bar downstairs, no two days are the same. So I can give you a more specific response, what role are you interested in?”
- Candidate: “Marketing Strategist”
- Bot: “Awesome, thanks. Typically, our strategists can be found studying data (nerds), developing campaign tactics, or helping the creative and account teams stay on-strategy.”
- Candidate: “Cool. Who do they work with?”
- Bot: “When not alone in a corner with their headphones on, strategists collaborate with people across the agency—most often with account managers and creative directors.”
- Candidate: “How do I apply?”
- Bot: “You can apply to be a strategist by…”
With conversational VUIs, the interactions are typically longer than a couple turns and cover a variety of topics. More sophisticated conversational VUIs leverage user input to tailor the interaction. In the example above, the bot asks for input (“… what role are you interested in?”). This enables it to provide a more detailed response. Later in the conversation, it uses the stored input (“Marketing Strategist”) to provide specific instructions on how to apply.
Agents don’t have to be sophisticated to carry a conversation. But, remembering inputs from earlier in the conversation help make the experience more seamless.
Making Features Discoverable Through Effective Prompt Design
Whether command-and-control or conversational, the VUIs need to help users understand how to interact.
They can do this by providing prompts. Prompts give direction to the user about how to interact with the VUI. Prompts that are most helpful to users:
- Are tailored to the input that is expected from or was most recently given by the user
- Set the user’s expectation as to what they can say and get in return
- Coach the user on how to say the input
- Cater to the context, as defined by the location, time, or purpose of use and the experience of the user
Providing the user an example of the expected input is helpful. This is called “coaching.” For example, the Centerline chatbot could say, ‘Interested in a particular job? You can ask what skills are required in order to get hired.” Even for narrow (or “constrained”) inputs, examples are key to a smooth experience. Another example – a chatbot could say, “What is your date of birth? For example, you can reply ‘December 21st, 1990.’”
Throughout the interaction, it’s important to manage the users’ expectations of the chatbot. Notice that the above example prompts focus on working at Centerline. Hopefully, this prevents users from asking questions about specific Centerliners, our clients, or other topics that would result in an error response: “I’m sorry, Dave. I’m afraid I can’t do that.”
Finally, helpful prompts are tailored to the context of the interaction. By thinking about who, what, where, when, why, and how the VUI will be used, you can produce a more elegant experience. For example, if our chatbot was headed to a job fair at one of the local universities, we might tweak our prompts to focus on topics of interest to prospective interns and grads (i.e. “bar,” “free food,” and “parties”). In the case of our recruiting chatbot, users will always be new. But, if you’re building a chatbot for repeat users, you’ll want to shorten prompts and tone-down the coaching as your users gain more experience.
To sum it all up, effective VUI experiences are intuitive. To get something done or carry a conversation, all the user needs to do is “just talk.” The designer/developer have studied the user’s expectations and the context of the interaction (we’ll talk about how, later) and used those insights to define a style of interaction and prompts that enabled a natural interaction.
Next, we’ll take on the remaining components of VUI interactions. Questions? Comments? Concerns? Feel free to reach out on Twitter – I’m @JaredBrickman