Best practices in conversation design
Conversation design is booming. People around the world are entering this new exciting field to help organizations develop human-centric conversational experiences for chatbots and voice assistants. With all these people entering the field, it’s important to start looking at some best practices in conversation design. Although the field is booming now, there are actually people that have been designing conversations for years.
Many of these experienced designers helped us develop course materials around conversation design. These best practices have proven themselves again and again and they are being used by the best conversation design teams in the world.
Let’s discuss a few of them. In this article, we discuss the following best practices in conversation design.
- Clear welcome message
- Onboarding dialogues
- Personality content
- Discourse markers
- Jenga Technique
Clear welcome message
You only get one chance to make a first impression. It’s the same with chatbots and voice assistants. When someone engages with your voice assistant or chatbot, your welcome message is the first thing people hear or read. It can make or break the experience. When you write the introduction message, a few things are important.
Make sure to:
- be likable
- design fallbacks
- be user-centric
- don’t pretend to be human
- guide the user
You want people to understand and enjoy the experience straight away.
You see, when you don’t do proper framing in the opening message, then you are making yourself vulnerable. When you don’t frame how you can be of value, then people can start asking you whatever they want and you often won’t know the answer.
Using proper framing increases the chances of your AI Assistant actually understanding the question.
Onboarding dialogues help people find their way with the AI Assistant. You see, not everybody has experience talking to chatbots and voice assistants yet. For those people, you want to have onboarding dialogues that allow them to ask questions about the concept of AI Assistants.
It should explain to people some of the following questions:
- what is an AI Assistant
- how should you talk to it
- what can it help with
- what does it do with data
- how does it store data
- how secure is it
A little personality goes a long way. It might seem silly to add jokes and chit-chat to a chatbot or voice assistant, but it actually has a serious purpose. People will want to test it out. They will ask it silly questions and jokes. These opportunities allow you to build a relationship with these people. The emotional connection is established via personality content.
This emotional connection comes in handy when your AI Assistant fails. When it doesn’t know the answer to a certain question, which can always happen, then people will be more forgiving because they have established a relationship with the chatbot or voice assistant. It’s, therefore, good practice to add personality content when you deploy your AI Assistant.
An acknowledgment signals that you’re receiving someone’s input. People use acknowledgments all the time. Each time you nod your head, say OK, sure, got it, and similar things, you’re actually acknowledging that you received someone’s input. You heard what they said.
An acknowledgment doesn’t commute any information. It just tells someone that you are tuned in. They are incredibly important in conversation design and particularly when you’re designing for voice assistants.
When someone says something, you want to signal that you received their input. Also, when you lead with acknowledgment, it gives people time to open their ears for the more critical information that is coming right after.
Examples of acknowledgments are:
- Got it
After the acknowledgment, you continue with a confirmation.
You’ve started your response with the acknowledgment; now it’s time for the confirmation. The confirmation repeats the input so that people know you understood them correctly. There are two types of confirmation. There is implicit confirmation and explicit confirmation.
We use implicit confirmation when we’re pretty sure we understand what the person said. In natural language understanding, we call this the confidence score calculated by your NLU engine. We repeat what the other said, and we continue straight away. We don’t ask him to confirm explicitly.
We use explicit confirmation when we’re not as confident that we understand what our user said. When our confidence score is low, we want the user to explicitly confirm what he said so that we can continue the conversation accordingly. It’s less efficient since it adds an extra step. However, the extra confirmation does build trust and can also be used when conversations get emotional or when discussing money.
You’ll notice that all our examples end with a question. This makes it a clear prompt that tells people what to do. In most cases, we want them to answer a question or click a button. This prompting has to do with turn-taking.
When you’re talking to people, it’s often pretty clear who’s turn it is to speak. Some people like to barge in whenever they feel the urge to talk. Others wait for the other person to stop talking and then reply.
Deborah Tannen has written about this extensively. She describes this as different conversational styles. Different people have different conversational styles. One is not necessarily better than the other.
However, in conversation design one is definitely better than the other. We want to signal people when it’s their turn to speak so that our chatbot or voice assistant doesn’t get confused.
We do this with proper turn-taking. We ask simple questions that are easy to answer. This keeps the conversation going. You can think of turn-taking as the gearbox of conversations.
We don’t want to answer a question and kill the conversation.
We prefer to have better turn-taking to keep the conversation going. This speeds up the conversation but it also makes people more comfortable. Each time they have to ask a question, it actually causes them stress.
Another concept part of best conversation design practices is called discourse markers. Discourse markers are like navigation in conversation experiences. You see, when people talk they mark or upcoming words or phrases relate to the previous discourse. Discourse markers are what binds together a piece of writing. It makes the different parts stick together.
Discourse markers can’t stand on their own. They always tie together multiple pieces of information. Removing discourse markers doesn’t change the truthfulness of a sentence, it just makes it harder to read.
Discourse markers can be used for different purposes. The can be used to:show cause and effect
soften what we say
change the subject
Let’s look at an example where we use discourse markers to organize information. This is what it looks like without discourse markers.
We serve burgers and fries, with that you can enjoy a free lemonade. We serve cold beers.
And this is what it looks like with discourse markers. You can quickly tell that the information becomes much easier to digest.
We serve burgers and fries, and with that you can enjoy a fresh lemonade. We also serve cold beers.
When you design conversations, it’s always helpful to do the one-breath-test. It’s a simple technique that helps you write better dialogues. It means that whenever you can’t say a message in one breath, then it’s probably too long.
When people talk to your voice assistant, it’s difficult for them to remember everything that the assistant says. They have a short attention span and since there are no visual clues, it’s tricky to remember everything. That’s why you want to keep messages short and simple.
The one-breath-test helps you do that. After every conversation that you design, do the one-breath-test and identify messages that you can shorten using the Jenga Technique.
Reading takes effort. It takes 250 milliseconds for your brain to process a word. That means it takes 25 seconds to read 100 words. That’s should already inspire you to keep your messages short and simple. Here’s the good news. On average, you can reduce every message by about 50 percent. That means you can take out about half of the words.
The technique is named after the game Jenga. In that game, players take out one brick at the time until the tower collapses. In conversation design, your goal is to take out as many words as possible without the tower of meaning collapsing.
In this extensive article on conversation design, we discussed some of the best practices that you can use as a conversation designer when you’re working on AI Assistants, like chatbots and voice assistants. (33 words)
In this article, we discussed best practices for conversation designers. (10 words)
We took out about 70 percent of the words and the tower of meaning still stands!
Final thoughts on best practices in conversation design
Now that companies are reaching for AI Assistants to automate 85% of conversations with customers, we’re seeing more people entering the field of conversation design. This is really exciting. It offers millions of people career opportunities and a great learning experience.
There are many best practices that you can use as a conversation designer. We listed some of them above, but there are many more.