As a subset within User Experience, Conversation Design considers not only how your customers will interact with your messaging or voice bot, but also what happens behind the scenes to make their experience frictionless, conversational, and intuitive.
As you begin on your conversation design journey, keep these best practices and principles in mind:
Design Natural, Intuitive Conversations
When thinking of how your virtual assistants will converse with your customers, it's important to recognize the overall tone of the conversation. People don't talk like robots, so having a robotic conversation inherently feels unnatural. So, why not make sure your bot talks more human instead?
Turn-taking, Reciprocity, and the Cooperative Principle
In general, conversational copy should be brief, and require low cognitive load. Use progressive disclosure, providing just enough information for the conversational turn.
Examples:
🚫 "It looks like your order was placed on Monday the 1st of August, at 2:47pm from an IP address located in Ohio. Does that sound right to you?"
✅ "Your order was placed on August 1st, right?"
With this in mind, you should also re-adapt long website FAQ copy for conversational interfaces.
Note: Most conversational use cases require that your bot assumes the onus of progressing the conversation, rather than the user. This is particularly important when leveraging Generative AI.
Be Mindful of Channel-Specific Actions
Certain phrases may not be as all-encompassing as you may think.
Examples:
🚫 "Click the link below." (Does not apply when using a touchscreen device)
🚫 "Tap the button to continue." (Does not apply when using a desktop device)
✅ "Select the link to continue."
✅ "Choose from the following options."
Note: For Voice bots, follow the “Rule of 3” when listing choices, and end your dialog turns with a direct question or call-to-action. Voice commands should be simple and intuitive.
Utilizing NLU and GenAI
Determine if you will use NLU, Generative AI, or both to power your conversational AI’s understanding. With this in mind, plan for how you will handle small talk and other commonplace conversational situations/commands:
- Hi, hello
- Thank you, thanks
- Are you a human?
- Can you repeat that? Come again?
- Go back, main menu, start over
- End, Close, Stop
- Agent, I want to speak to a human
- Hold on, wait
Some additional situations may include…
1️⃣ Are there any highly sensitive topics you will auto-escalate to a human immediately?
2️⃣ How will you handle repetitive abusive sentiment or pranksters?
3️⃣ Inactivity / Abandonment
Note: Be on the lookout for any potential looping that could occur in your conversation design, and remove it. You wouldn't want your virtual assistant to repeat itself over and over again just because someone says "thank you". Also be sure to check that error messages are written in a helpful manner, enabling users to progress and get back on track.
Set Clear Expectations & Bot-to-Human Handoffs
Now that we know some general best practices of conversation design, how can we put them into action? Let's start at the top of the conversation!
Defining a "Welcome"-ing Experience
When starting the conversation with your customers, always be sure to introduce your virtual assistant and set expectations of what it can and can’t handle.
Examples:
🚫 "Greetings. How can I help you?"
✅ "Hi there! I'm Facto, your virtual assistant. I can answer some of our most common questions and even check your order status automatically. So, how can I help?"
Note: Do not pretend to be a human. As nice as it is to have a human-sounding bot, you don't want to deceive your customers.
Defining How a Conversation Ends
There are a few different methods to ending a conversation, but you should always want to make sure the customer has nothing else they'd need assistance with. Never simply provide an answer to a question then close the conversation, so be sure to clarify with the customer if anything else is needed.
You will also want to define how and when you’ll offer a Post-Conversation Survey. Ask yourself these key questions to help determine this:
- If the virtual assistant couldn't provide information nor escalate to a live agent, is a Post-Conversation Survey necessary?
- Will you ask the same survey questions for if the customer only spoke with your virtual assistant vs. a live agent-handled conversation, or will the survey questions be different for each scenario?
Get Your Safety Nets Ready
Virtual assistants can only handle so many scenarios, so you'll need to map out a plan for Fallbacks and Escalations.
If the customer's utterance is unrecognized by your bot, use Fallback and re-prompt strategies to handle unrecognized inputs, API errors, and long processing times.
Examples:
Virtual Assistant: "What is the order number you're trying to track?"
Customer: "123ABC"
🚫 Virtual Assistant: "Sorry, I didn't get that. Try entering this again." (Does not provide any follow-up/why this is unrecognized)
✅ Virtual Assistant: "Sorry, your order number should be 8-digits in length and only contains numbers. Can you please provide that again for me?"
Note: Plan for potential escalation to live human agents when a query cannot be resolved or if it needs further assistance to complete. You will also want to decide when you will auto-escalate vs. asking the user to confirm before escalation. If the customer continuously types "agent" to "human", an auto-escalation may be necessary, but for human-error cases of mistyping some requested information, we may want to confirm if they'd rather speak with a live agent first.
Design with Context
Consider how your design will adapt to context across these five areas:
1️⃣ User - How will information and conversations flow for new vs. returning customers
2️⃣ Conversation - What procedures and steps will be repeat with every conversation?
3️⃣ Bot - Is there any information that the bot should keep track of that is not related to the user or the conversation?
4️⃣ Situational - What procedures and steps may or may not be presented with each conversation?
5️⃣ Global Contexts - Is there any information that should be shared across all conversations and contexts? (Live agent hours of operation, etc.)
Note: In Conversation Builder, you should also leverage dialogs as reusable components. Don’t create duplications of the same information, you can simply reference existing dialogs if they apply to multiple contexts. You should also organize your flows in Conversation Builder with a numbering and naming system. (ie: 0- Welcome, 1.0 Main Menu, etc)
Simplicity over Complexity
When writing copy for your virtual assistant, consider your customer’s mental model. Remember that customers may not be familiar with the terminology of your business, so you should avoid jargon and acronyms when possible.
Examples:
🚫 "So, what is your PR-Number?"
✅ "To verify your account, can you please provide your Platinum Rewards Number? You can find this on your Account page by clicking the icon on the top-right of our website."
Note: Use similar words and phrases that your customer does, especially when designing menus, phrasing dialog prompts, and building out intents and training phrases in Intent Manager.
Design Alignment & Collaboration Tips
Lastly, here are some tips to help your conversation design grow with proper collaboration and communication across roles:
🌟 Identify team members responsible for Conversation Design and Development work. This will ensure that everyone's roles and responsibilities are clear and who needs to communicate with whom for bot design and development maintenance.
🌟 Ensure proper stakeholders across your company are involved and aligned with your conversation design strategy. Conversation designers often must collaborate with Product, Engineering, Contact Center teams, Operations, and Marketing in order to make the right model, so identifying these connections as early as possible is key!