Conversational design for chatbots is key to making it engaging and useful for users. Learn about the key principles and best practices to deliver human alike conversations.
Designing a conversational AI is more than just building the best AI models that deliver the most factually accurate response to user queries. It requires a sound understanding of how humans communicate and the skill to be able to replicate that in conversations with bots.
A good conversational AI will not only be efficient in providing accurate responses but it will also have a helpful, empathetic personality that makes humans want to interact with it. To design a chatbot that can effectively converse with humans, it is ideal to tap into expertise in technology, psychology, copywriting and even linguistics. This is essentially the scope of conversation design.
What is Conversation Design?
Conversation design is the process of designing interfaces and interactions between humans and machines that mirror our real world conversations. It is a broad subject that is applicable to websites, mobile apps, chatbots or indeed any interface between humans and machines.
In the context of chatbots, it involves considerations around the personality of the chatbot, the user experience it provides, multiple scenarios in a conversation and the end user’s goal among other factors.
Psychology of Conversations
The task of designing good conversations can often be taken for granted, since our day to day conversations are often effortless, especially if we speak a common language with others around us. But language is one of the most complex aspects of human evolution. And designing bots that can converse with humans is even more difficult.
To get anywhere close to an effective conversational AI, one needs to turn to psychology for insights, especially the work of philosopher Paul Grice.
The Cooperative Principle
Paul Grice was a British philosopher of language, whose work influenced the study of semantics in a big way. He developed the Cooperative Principle to describe how people interact with each other. He stated the principle as follows:
“Make your contribution such as it is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged”.
The principle is called so because listeners and speakers must speak cooperatively and mutually accept one another to be understood in a particular way. It is not meant as a prescriptive command but rather as a description of people’s normal behaviour in conversations. It essentially outlines the underlying principle that what we say in conversations furthers the purposes of these conversations.
|PRO TIP: When designing chatbots, keep the cooperative principle in mind. Have it work together with the human, find common ground and move the conversation forward towards a goal.|
The cooperative principle can be broken down into four points, known as maxims, which serve as guidelines to follow in developing good conversations that get meaning across.
The 4 Maxims of Conversations
The cooperative principle comprises of four maxims which explain the link between utterances and what is understood from them. These are the Maxims of Quality, Quantity, Relevance and Manner.
Maxim of Quality
The maxim of quality refers to provide information that is true. Being truthful in a conversation means avoiding falsehoods and saying things without adequate evidence.
In the case of chatbots, this means ensuring that the bot only provides responses that are based on facts and truth.
Maxim of Quantity
The maxim of quantity means that we provide as much information as needed to achieve a goal, no more and no less. Think of the difference between having a conversation with someone who replies in just Yes or No statements and another one with someone who goes on a long monologue with unnecessary detail. This is a critical principle that chatbots must follow in order to continue the conversation with a user and keep them engaged.
Consider the following two examples of a chatbot BankPal on a bank’s website.
User: When does the nearest branch open?
BankPal : 9 am.
User: When does the nearest branch open?
BankPal: There are 4 branches for XYZ bank located within a radius of 5 kilometres. The nearest branch for XYZ bank is 250 meters away at 23 High Street and it opens at 9:00 am.
Neither of these responses are ideal. A better version that provides just the right amount of info would be something like “The nearest branch is at 23 High Street and it opens at 9:00 am.
Maxim of Relevance
The maxim of relevance suggests that we engage in conversations that are relevant to others. For chatbots, this would mean replying to queries with information that help the user with their goal. A response that is not relevant to the user can break the natural flow of conversation and turn the user away.
Consider the following example of an insurance chatbot InsurPal, helping a customer with a query.
InsurPal: Would you like to book an appointment with our agent?
User: When is he available this week?
InsurPal: Please answer yes or no.
InsurPal: Would you like to book an appointment with our agent?
User: When is he available this week?
InsurPal: Your agent is available 9 am to 12 noon on Wednesday and Friday.
The example on the right is more helpful for the user while the one on the left did not have the bot respond with a relevant response.
Maxim of Manner
When designing conversations, it is also important to follow the maxim of manner. This means being clear, brief and orderly while avoiding obscurity or ambiguity.
For example, when an insurance bot responds to a customer completing a purchase, it could respond in either of the two ways.
InsurPal: Your transaction has been completed.
InsurPal: Great, you’re all set!
Clearly, the response on the right is a better and more natural way of closing the conversation. It is clear, brief and easy to understand. This maxim is also what contributes to the “personality” of the chatbot in a big way.
Two Main Types of Conversations
Chatbots and conversational AI systems have multiple use cases across industries. But no matter what the conversation is, it usually follows a standard pattern.
There is a traditional three-act structure that has been effective in storytelling from the time of ancient Greek philosophers, whether it be in books, plays or movies. It comprises the start (where the backstory is set up), the middle (where the conflict and obstacles to overcome are established) and the end (where the hero of the story comes to resolve the conflict).
Dialogues and conversations also follow a similar three-act structure and it is useful for designing good chatbots as well. Broadly speaking though, the types of conversations that chatbots are required to engage in can be categorised into two.
This is where user comes to the bot asking for help with a query or problem. The copywriting techniques to adopt for this type of conversation should be aimed at addressing the query or problem. The structure here involves beginning by recognising the intent, asking questions and confirming the solution. The bot should be able to introduce itself, manage expectations and show empathy. You also want to avoid falling into a loop trying to recognise the intent if it was not obvious at the first attempt.
Sales or Persuasive Conversation
This type of conversation is applicable when the bot is used to conduct a transaction, monetary or otherwise. In this case, it is the bot that often initiates the conversation and tries to persuade the user. For example, the bot may be tasked to persuade the user to fill in a form to capture them as leads, leave a review or make a purchase.
Here the three-act structure should start with the first task of sparking interest. For example, it may be in the form of a simple question to get their attention or catch their curiosity by presenting them with a unique benefit. This should be followed by attempts to persuade the user to make small commitments like clicking a Like or Upvote button and other soft calls to action.
Key Aspects of Conversation Design
With the four maxims of the cooperative principle, we have a set of basic guidelines to follow to design good conversations. For chatbots specifically, there are a few best practices that conversation designers follow.
Prompts and Bot Introductions
In introducing the bot to a user, you need to keep in mind the maxims of manner and quantity. A pleasant introductory question is good but it also makes a big difference if you can make it clear to the user what purpose the bot serves.
“Hi there, what can I do for you?”
Pleasant but it is still vague what the bot can do
“Hi there, I can help you with queries about our insurance products, compare our various policies, make a purchase, book an appointment with an agent and file for claims. What can I do for you?
Pleasant and informative but there is too much information which can cause cognitive overload for the user.
“Hi there, I can help you understand our insurance products, purchase a policy and more. What can I do for you?
Pleasant, informative and leaving open the option of “and more”. Users can then ask about these other options if they wish.
In the event that the bot is unfamiliar with the user’s choice or it fails to understand the intent, fall back messages can help avoid confusion. A simple “I’m afraid I don’t know about that. Let me transfer you to a live person” can help close the conversation in a polite way in this instance.
It is also important to design for frequency when writing dialogues for chatbots. This means allowing the bot to adjust the introductions for repeat visitors. Think of how as a user you would respond to hearing “Hi there, what can I do for you?” again and again multiple times over the course of a week if you frequent the website or app with the bot. This can turn off the user due to the seeming detachment and coldness. A good workaround for this is to have the bot detect visitors who return within the space of a month and adjust the intro to a more familiar “Hi there again, what can I do for you today?”
Natural Multi-turn Conversations
If you think about the conversations we have with other humans everyday, you will find that there are a number of quirks at play aside from the words and sounds we utter. The vocal tone, tilt of the head and body language are important non-verbal cues that supplement the words we speak and provide extra context.
The idea of turn-taking is a good example of this. When we engage in conversations, we often adjust our pitch or turn our head to let others know that it is their turn to speak. Think of the tone you use to end a question and wait for an answer. How do we apply this trait to bots when there is no visual or other non-verbal queues available from the user?
This is where ending prompts with a question makes a big difference in moving the conversation forward. Consider the following exchange.
User: What life insurance policies do you sell?
InsurPal: We have the ABC Basic, ABC Premium and ABC Savings schemes.
The conversation seems to end abruptly and depends on the user to ask the next question. Ending the reply with “Would you like to choose one to learn more?” will prompt the user to continue the conversation.
Acknowledgements, Explicit and Implicit Confirmations
Acknowledgments assure chatbot users that their input was received. “Sure”, “Okay”, “Alright”, “Excellent”, “I see” are all various ways that the bot can acknowledge the user input and make them feel that they are being heard. These also add a touch of humanity to the bot and builds trust with the user.
An explicit confirmation is when a bot asks the user to confirm by repeating parts of the query explicitly. It is useful for situations when the bot’s confidence in recognising the intent is not high enough or when the stakes are high. For example, if the tasks involve transferring large sums of money, or sending a message to a number of contacts it helps for the bot to check again to make sure there isn’t any confusion.
An implicit confirmation is one that does not require a confirmation from the user, but which also leaves the option open for the user to confirm or deny. It makes the conversation a lot more natural, closer to how humans talk with each other.
User: Can you book an appointment with the Cardiologist?
Medbot: You would like to book an appointment with the Cardiologist, is that correct?
The engine’s confidence is low, say below 65%
High stakes situation, such as when money is involved (financial transactions)
User: I want to file a claim.
InsurPal: Sure, I can help you file a claim. What is your insurance number?
The engine’s confidence is high, say above 65%
The user can correct if there is a mistake while proceeding
Prompt Clarifications With More Explanation
Yet another technique that humans use in communication which can be adopted to improve chatbot conversations is prompt clarifications. This involves repeating the question with more words to ease the user’s comprehension of the prompt. Let’s imagine a virtual assistant introducing itself.
- It might begin with “Hi there, what can I do for you today?”
- If there is no response from the user, it makes no sense to repeat the intro over and over again.
- Instead the bot can add words to improve and clarify the prompt like so: “I can help you find the right insurance product, file a claim or transfer you to your agent.”
- If there is still no response, more words can be added to help the user decide as such: “For questions about insurance products, press 1. To file a claim, press 2. To talk to your agent, press 3.”
It’s not easy to predict a user’s emotional state when they are interacting with a chatbot. They could be casually looking for information or frantically trying to make a transaction or get in touch with someone. Bots should keep this in mind and go beyond merely providing factually accurate responses. In customer service especially, half the job is done if the bot can empathise with the user.
Empathy is about showing that you understand the user’s situation. Bots can never be as empathetic as humans but their conversations can be designed in such a way that users feel that they are heard and empathised with. Words and phrases that show apology and gratitude are useful ways to display empathy.
User: I want to update my phone number connected to my insurance policy provider. I tried doing it in the app but the 2FA code was sent to my old number. This is frustrating!
InsurPal: I am sorry you are unable to update your phone number through the app. Would you like to reach out to our call centre?
The above response includes an apology and a confirmation. Note that it does not repeat what the user is feeling by saying “I am sorry. That must be frustrating”. This might appear as if the bot was trying too hard and could backfire.
User: Ok I would like to renew my policy for another year.
InsurPal: Excellent. Thank you for continuing to choose our policies. Let me help you with the renewal contract.
These are just some of the ways in which bot conversations can be designed to show empathy. It is also important not to create responses by assuming too much about a user’s situation. Consider the following example.
User: I want to change my address to my new home at #12-21, 23 High Street.
InsurPal: Congratulations on moving house. Let me update your address in our records.
The change of address could have been due to many reasons, including personal and family issues, maintenance and repair problems or even disasters. A response like the above from the bot can appear tone deaf and ruin the relationship.
Usually, the most common aspects of a chatbot people talk about are the technology and language recognition, efficiency in automation and how factually accurate the responses are. But these attributes alone are not enough to make a chatbot great. Good conversation design is essential to ensuring that users keep interacting with the bot, find the experience useful and engaging enough to come back to it without bypassing it to speak to a live agent straight away.
To find out how KeyReply can help you design engaging and useful chatbots for your organisation, talk to our experts today.