Welcome to the second post in our series on chatbots and emotional intelligence. Last week, in Part One: Shortcuts to Chatbot Emotional Intelligence , we covered why it’s important to have a user-centered design and strategy process in place before you dive into creating conversational flows for your bot. Today, we’ll discuss five user-centered design considerations that can help you breathe life into your chatbot.
Chatbot Design Focus One: Usefulness
Building something that’s going to be useful should be our top concern. We don’t want to go building Skynet , do we? No, we want to build Johnny Five! Because Johnny Five is our friend and wants to help us.
So where do we place our focus within such a broad landscape of possible uses? Try to think about situations where a bot’s computational powers would be handy. How could a bot help us more than the average human with a smartphone?
“Try to think about situations where a bot’s computational powers would be handy. How could a bot help us more than the average human with a smartphone?” When deciding which conversation flows to include, look for tasks–like sifting through huge datasets–that are generally hard for humans to do, but easy for bots.
For example, we might ask a bot…
“Which restaurants would all of my friends enjoy?”
“What time fits everyone’s schedules?”
Or even something like…
“Which restaurants have at least four stars on Yelp and vegan and gluten-free options, with a table for six available at 7:30?”
We’d end up hangry trying to come to a consensus—but, a bot can quickly make sense of the mountains of data required to make the optimal decision.
Chatbot Design Focus Two: Helpfulness
Similar to usefulness—but no less important—is helpfulness, the ability to anticipate a need and offer ways to satisfy that need in a timely manner. Mobile technologies like location services and beacons make many just-in-time interactions possible. Consider including these use cases in conversational flows but be sure to include an alternate path for desktop users.
In terms of the chatbot’s GUI, card-based UI elements have become increasingly popular with the rise of Google’s Material Design. That’s fortunate for us because cards provide a convenient way to serve up interactive chunks of information, combining text and graphic elements to create a more user-friendly experience.
Prompts are another way our bot can help users when the conversation gets into uncharted territory. A simple, “Sorry I didn’t get that—try asking me about tech news or headlines,” will go a long way towards keeping our users from getting frustrated.
In summary, usefulness is defined by what problems the bot solves for and helpfulness is how it helps solve for them.
Chatbot Design Focus Three: Flow and Cadence
Mimicking the flow and cadence of a conversation with another human can help users get used to the idea of chatting with a bot. To create a natural flow, shorter messages invite banter and feel more conversational. We can include a realistic lag between messages with a typing indicator to allow your user time to process the chatbot’s responses. This also gives the cadence of the conversation a more authentic feel. First impressions really do make or break an interaction, so we should consider our greeting carefully. It’s often helpful to explain what our bot can do in the first interaction, but it can be jarring for the user to get a huge chunk of text before exchanging a greeting. Strike a balance with a succinct salutation. Let the user make the first move.
Sign-offs are just as important. In a real conversation, we typically taper off the length of our messages as the conversation wraps up. Our chatbot should do the same. Remember that it may be a few days or weeks between message conversations, so it’s helpful to sign off with a prompt for next time:_ “Sure thing! If you need anything else, just tap one of these options.”_
Chatbot Design Focus Four: Humor
Personality is what makes a bot (or human for that matter!) enjoyable to interact with. I’ve even heard that comedy writers are becoming the next hot UX hires in the hopes of making copy more engaging. We often use humor to lighten up a situation and smooth over our mistakes. We might consider using this tactic when the bot encounters a limitation or needs to give an error message. In many cases, responding in this way will smooth over frustration and encourage your users to try to get another funny response, driving more interaction. Users understand that bots aren’t infallible, so addressing that fact with some self-deprecating humor can be a quick way to win your user’s sympathy.
To succeed with humor, consider what type of personality your users would find engaging. Make sure to fit the tone to the task at hand, as well as the brand your bot is representing. For example, if you’re building a banking bot, avoid flippantly humorous error messages when a user’s balance can’t be retrieved. For example, “Oops! I couldn’t find your money. Why don’t you try checking the last place you put it?” won’t go over well. This is a very important place to check in with our User Personas to get a feel for their preferences and their intents. This example is from Luka , which started as a restaurant suggestion app but has added personalities and functionality since its initial launch. Here, we can chat with Erlich Bachman from the show Silicon Valley, who is all personality and puns. Luka shows just how engaging a personality can be—so much so that it can function as the sole utility of the bot.
If humor doesn’t work, sentiment analysis is the next step. Sentiment analysis uses Natural Language Processing (NLP) to allow a bot to judge the user’s tone and react accordingly. If someone is becoming increasingly frustrated, the bot can recognize that and adjust its language to try to placate the user or alert a customer service rep that a human may need to step in. Sentiment analysis is by no means an easy lift. Just keep trigger words or phrases in mind that you might be able to address when training your bot’s intents. Think phrases like…
“what the hell?”
“You’re joking, right?”
Chatbot Design Focus Five: Intelligence
Bots are still far from the uncanny valley or any type of Her scenario. It would take years to develop a bot that handles anything we can throw at it. Instead of aiming that high, we typically choose a few conversational flows and get them right before scaling.
Some form of statefulness is a good goal. This requires us to establish conversational context, which is the ability to correctly match up pronouns with their proper subject between interactions. Here is an example.
“What’s the weather like in Brooklyn?”
“Cloudy and 77.”
“Will it rain there tomorrow?”
Another form of statefulness is preferential context, or saving and revising user info like name and preferences. These abilities help streamline interactions so users can take advantage of conversational shortcuts and aren’t starting from a blank slate each time.
Ideally, we should plan for our bot’s shortcomings by acknowledging its limitations and handling them gracefully. A hybrid approach to customer service interactions, combining human and bot interactions, will help our users avoid frustration. It’s important to avoid awkward handoffs between bot and humans in these cases.
The example below isn’t the best in that regard. A better solution would be for the bot to introduce the new voice in the conversation and even show a different name and photo to avoid confusion. Something like…
"That’s a bit beyond my abilities. Let me get someone who can help you. Be back in a sec!”
“Here’s Jill from Member Services. She can help you update your account information.”
Conclusion - Testing Your MVP
That’s all five of the most important considerations for Bot UX design. How does your bot measure up? Well, that’s up to your users to decide.
Now that you have an idea of your conversational flows, this is a great point to validate your assumptions with some user testing. If you haven’t built and trained your bot yet, there are a few ways to fake chatbot interactions. You can depend on your users and the complexity of interactions to define the fidelity of prototype to test with. Some options, from high to low fidelity, include:
- Clickable prototypes including each conversational interaction
- Scripted interactions with a researcher posing as the chatbot
- A Twine prototype
- Tree testing
Look for places where users get frustrated or go down the wrong path when trying to complete an assigned task. Then you can easily iterate language and test alternative flows to improve the user experience and create bot-and-human harmony.
Check back later for part three of this series where I’ll walk you through best practices for documenting your conversation flows.