Thoughts

Design conversations to be helpful, not human

Braden Ream
|
January 15, 2021

Human conversations

Throughout my time at Voiceflow, I’ve seen and worked with a lot of conversation design teams. One trend I’ve noticed, especially with designers just getting into conversation design, is focusing too much on making conversations sound human. I’ve seen designers focus their time on small talk, jokes, and over-the-top personas, all to create a more human experience.

As conversation designers, our goal is not to design human conversations; our goal is to create helpful conversations that help users achieve their goals.

Small talk in conversation design

When humans converse with each other, we follow the cooperative principle, which states that both people work together to achieve an objective through conversation. This objective could be as lighthearted as telling a story or transactional as buying a car. The foundation of these human conversations is built on empathy or the mutual understanding of each other’s desires and emotions. When two people have a relationship, i.e., empathy, conversations are more fluid and less rigid as both parties know the other is empathetic toward their needs. This base of empathy is why we, as humans, lean on small talk like “how’s your day going?” — although we don’t always care how the other person’s day is going, we leverage this sentence to showcase our empathy toward other humans.

Unlike humans, computers are not empathetic (yet), and because of this, assistant small talk is often considered invaluable or unnecessary. When I talk to my Starbucks assistant, I don’t want to tell the app how my day is going — I want a coffee. Unlike my barista, I know that the assistant is incapable of being empathetic, so expecting an emotional base, the same question would add friction and unnecessary steps to my order.

Furthermore, today’s NLP/NLU technology is far from perfect and can only manage well designed structured conversations. Small talk is the prime example of an unstructured conversational element that can only serve to trip up and break assistants, or worse, be contextually irrelevant or bland.

For example:

Coffee bot: “How are you doing today?
Customer: 
“I’m actually going through a really rough day,”
Coffee bot:
 “Great! What can I get for you?”

Yikes. The perils of blanket responses to unstructured prompts.

When designing conversations, ditch the small talk. While it may be an impressive tech demo in the board room with perfectly planned utterances, it’s not always going to match your users’ expectations or context.

Tangential functions

When designing an assistant, it’s essential to design with a goal for the user in mind. If you’re designing a pizza ordering assistant, you shouldn’t have to design flows for Intents like “tell me a joke.” These flows fall into the “pointless small talk board room demo” category as they are not critical to the user’s experience. Instead, these side tangents often bloat the experience and hide what’s really important. If your Intent analytics show that users are consistently asking your Pizza design for a joke, so be it, but otherwise, make sure to focus on core flows/experiences.

Human personas

Conversational personas are important, but designers need to remember that in many cases, these personas represent the brand as an assistant across all sorts of situations and environments – not necessarily the brand as a human. Although there are several engaging ways to leverage human-like characteristics and humor in conversation design, they aren’t always necessary. For instance, in games, education, or perhaps even brand creation, using voice actors, custom rhetoric, or personality can drastically improve a conversational experience. In contrast, transactional experiences such as banking, POS system ordering, or logistics may not leave room for such liberties.

In these situations, added small talk could slow down the conversation and increase the time it takes to complete the user’s goal. In many transactional use cases, users know they are talking to an assistant, not a human, so any dialog unrelated to their objective could be considered a bad experience.

Wrapping up

When designing conversations, remember that your goal is to create a pleasant conversation with a computer that achieves an objective. In most cases, your goal is not to pass the Turing test and foster the most human conversational AI out there. Therefore, when designing transactional or utility-focused conversations, avoid small talk, tangential functions, and overly human personas and focus on building intents for their core needs/goals.


Comments? Please send me an email: braden.ream@voiceflow.com. You can also subscribe to our blog to stay up-to-date with the latest Voiceflow news.